Absolute lymphocyte counts (cells/μL) (CD3+, CD3 +CD4+, CD3+CD8+), B cells (CD19+) and NK cells (CD56+/CD16+) and percentages of CD8+/HLA-DR+/KI67+ T cells were determined from heparinized whole blood using standard flow cytometry methods. IL-18, CXCL10, GZMA, CEA and CA27–29 were analyzed in plasma using luminex (Myriad-RBM) and ELISA assays.

Tumor-infiltrating ICs (lymphocytes, macrophages, dendritic cells and granulocytes) as detected by hematoxylin and eosin (H&E) staining were scored as a percentage of the tumor area comprised of tumor cells and desmoplastic stroma [7]. PD-L1 expression on immune cells and tumor cells was evaluated using the VENTANA SP142 immunohistochemistry (IHC) assay (Ventana Medical Systems, Tucson, AZ, USA) [9]. Immunohistochemistry was centrally performed (HistoGeneX, Antwerp, Belgium) using C8/144B (CD8) and MRQ-26 (CD163) antibody clones (Dako, Glostrup, Denmark). Aggregated data for these biomarkers from the TNBC cohort in the PCD4989g study [7] is provided in Additional file 1: Table S1.

Gene expression levels were quantified by TruSeq RNA Access RNA sequencing (Illumina, CA, USA) [10, 11]. For comparison, the data for the individual tissue samples are displayed along the aggregated values from the rest of the PCD4989g TNBC cohort (Table S1, n = 103) [12]. TNBC molecular subtypes were assigned as previously described [13], with modifications to adapt for the use of formalin-fixed, paraffin-embedded tissues obtained on our study. Mutation detection, tumor mutational burden, somatic/germline status and clonality were assessed using the FoundationOne® platform as previously described (Foundation Medicine, Inc., Cambridge, MA, USA) [14–16].

A 48-year-old woman with a long history of TNBC (Fig. 1) was enrolled in the Phase 1a study of single agent atezolizumab (PCD4989g; NCT01375842) on March 5, 2013. In 1986, she presented with an early right breast cancer negative for the estrogen and progesterone receptors. Initial management included lumpectomy and radiotherapy, and then two chest wall recurrences were treated with surgery and adjuvant chemotherapy (1993). In 2009, the patient presented with another early right TNBC treated with surgery followed by chemotherapy with docetaxel + cyclophosphamide (T + C). She then developed skin nodules and left axillary lymphadenopathy and received palliative capecitabine. In 2010, she developed progressive mTNBC involving the sternum and mediastinal lymph nodes and received 9 cycles of gemcitabine, carboplatin, and iniparib (GCI) until summer 2011. In late 2011, lymph node recurrence was treated with axillary lymph node dissection (2/5 lymph nodes involved with tumor) and radiotherapy. In 2012, she developed progressive disease (PD) and received 3 cycles of a whole cell breast tumor vaccine, trastuzumab, and low dose cyclophosphamide (NCT00971737). Upon progression, she was enrolled in the Phase 1a PCD4989g atezolizumab monotherapy trial, first dosed on March 11, 2013. She had a partial response (PR) by Response evaluation criteria in solid tumors (RECIST) and Immune-related Response Criteria (irRC) after 4 cycles. On May 31, 2013, the patient had an irPR per irRC and unconfirmed PR per RECISTv1.1. In July 2013 she experienced a pseudoprogression: appearance of a new nodal lesion (PD by RECISTv1.1) while still responding according to irRC. She was clinically well and continued treatment. On November 2013, the patient progressed by both RECISTv1.1 and irRC with the appearance of two new nodal lesions. On January, 2014, her target lesions were still in response (Fig. 2a), while the new nodal lesions were enlarged but stable. On February 2014, after 16 cycles, she discontinued atezolizumab exposure per protocol, with close surveillance.Fig1Fig. 1

Clinical course of disease and time-points of collected tumor biopsies. On the right: chronological schema of disease appearance and treatments, on years. On the left: time of sample collection (red dots) on days respect to initiation to atezolizumab first exposure. CMF: cyclophosphamide, methotrexate and fluorouracil; Biopsy Dx: diagnostic biopsy; LAD: lymphadenopathy

The impact of atezolizumab on surrogate plasma biomarkers of breast cancer progression (CA27–29 and CEA) and T-cell activation was evaluated. Plasma CA27–29 decreased prior to the first radiologic response (+63d) but increased prior to PD (+183d) and then normalized (+245d, Fig. 2c), suggesting a progression between two response episodes. Consistent with a systemic pharmacodynamic biomarker effect of atezolizumab [9], CD8 + Ki67 + HLA-DR+ cells increased transiently after one cycle of atezolizumab (23d, 7.8-fold), followed by a nadir, then a slow rise over one year of treatment (Fig. 2d). Natural killer (NK) cells and CD8+ T cells followed a similar pattern, but the change was less profound (2- and 1.4-fold on +23d, respectively) (Fig. 2d). Similarly, interleukin-18 (IL-18) and CXCL10 (cytokines/chemokines induced by IFNγ), and extracellular granzyme A (GZMA, produced by cytotoxic T lymphocytes and NK cells) peaked after one cycle of atezolizumab and then returned to baseline (Fig. 2e). Although these changes in peripheral blood biomarkers over time have not been associated with atezolizumab clinical activity, they indicate a systemic but transient atezolizumab-induced systemic T cell activation [9].

Multiple tissue biopsies were collected over the patient’s clinical course, providing an opportunity to evaluate temporal changes in the TiME after successive therapies. PD-L1 expression on ICs and tumor cells, ICs by H&E, CD8+ T cells, and CD163+ macrophages was evaluated using IHC. Additional file 1: Table S1 serves as reference of for the median, range and interquartile values for the totality of the samples collected in the TNBC cohort of the PCD4989g clinical study [7]. Immune infiltration (ICs and CD8+ T cells) and PD-L1 IC expression were low in the early disease tumor specimens and increased after T + C and capecitabine exposure (−1176d to -556d, Fig. 3a and b). Over the course of further chemo/radiotherapy and breast tumor vaccine, ICs, CD8+ T cells, and PD-L1 IC expression remained elevated. Throughout atezolizumab exposure (+380d), ICs, CD8+ T cells, and PD-L1 IC expression were high, but decreased at relapse off atezo first exposure (+689d), most significantly for PD-L1 IC (10 to 1%) and to a lesser extent ICs (20 to 15%) and CD8s (8.5 to 6%). CD163+ M2 macrophages, usually associated with immunosuppression, first increased post-capecitabine, further increased post-GCI and XRT, declined after the tumor vaccine immunotherapy, remained low through the suspected pseudo-progression on atezolizumab (+380d) and peaked again at relapse off-atezo. The tumor had elevated immune infiltration and PD-L1 IC expression and peaked at relapse off-atezo (+689d, 6.23 to 25.69%). PD-L1 in the tumor cells was never detected (Fig. 3b). These results suggest that while progression while off atezolizumab had reduced PD-L1 IC, the tumor sample was still PD-L1 IC+ (≥1%) [3, 7].Fig3Fig. 3

Evolution of Tumor Microenvironment. Samples collected over time. Images at 100X (a) of tumor infiltrating immune cells (ICs), PDL1 in tumor cells and tumor infiltrating immune cells, CD8 T cells and CD163 macrophages were assessed by hematoxylin& eosin or immunostaining. b Quantification of the parameters in (A) are displayed as a percentage of tumor area and evaluated over time with respect to atezolizumab first exposure. c RNA-based signatures associated with T cells, regulatory T cells, CD8 effector T cells, NK cells, B cells, macrophages, immune checkpoints, cancer associated fibroblasts, cytolytic activity, antigen processing, angiogenesis, and proliferation were derived from RNA-Seq and plotted as PC1 scores and displayed over time. As reference, the aggregated value for the samples from the TNBC cohort in the PCD4989g study (PCD, all) is displayed as box plots representing median, 25th and 75th percentiles and the vertical bars represent range (maximum and minimum). d TNBC subtype classifiers were derived from RNA-Seq for each sample. Heatmap denotes relative RNA expression of genes involved in the subtype classifiers in the analyzed samples. BLIA and BLIS prob.: probability that the samples are BLIA or BLIS

To evaluate the evolution of the immune, stromal and tumor biology, RNA-based gene expression signatures were analyzed in a subset of tumor samples with evaluable material (−1834d, −1176d, −556d, +380d and + 689d) and compared to the aggregated levels for all the TNBC patients in the PCD4989g study (Fig. 3c). Consistent with the IHC findings, RNA-signatures associated with T cells, NK cells, antigen presentation, cytolytic capacity and immune checkpoints were low in the primary tumor and increased over time, prior to and while on atezolizumab. All of these signatures declined, except for macrophages, at the time of PD while off atezolizumab first exposure (+689d, Fig. 3c). Tumor stroma and angiogenesis have been associated with poor clinical outcomes in early TNBC [17, 18]. RNA-based signatures for cancer-associated fibroblasts (CAFs) were overall low, except of an increase post-capecitabine (−556d), but returned to lower levels post-atezo. Angiogenesis and proliferation gene signatures were overall medium-high throughout the clinical course of the patients, regardless of therapy. The RNA-based results support the immune IHC data and that off atezo relapse is not associated to overall loss of TiME.

TNBC is a heterogeneous disease composed of several molecular subtypes. Four distinct TNBC subtypes were identified by RNA profiling: luminal androgen receptor (LAR), mesenchymal (MES), basal-like immune-suppressed (BLIS), and basal-like immune-activated (BLIA) [13]. In early TNBC the prognosis is worst for BLIS tumors and favorable for BLIA tumors. TNBC subtype profiling of the tumor samples showed an evolution from LAR in the primary tissue (−1834d) to BLIA in the lymph node metastasis after TC (−1176d, −556d, +380d, +689d) (Fig. 3d). Post-TC (−1414d) and at PD off-atezolizumab (+689d), the BLIA samples had a significant BLIS component (38 and 42%, respectively), while the post-capecitabine (07/02/2011) and post-atezolizumab exposure (+380d) samples had a lower BLIS component (17 and 26%, respectively) (Fig. 3d). These fluctuations in LAR/BLIA/BLIS biology seemed consistent with the immunohistological and RNA variations described above.

Studies in bladder cancer, lung cancer and melanoma have shown an association between high tumor mutational burden (TMB) and response to immune checkpoint blockade [19, 20]. This patient’s tumor biopsies were subjected to comprehensive hybrid-capture-based genomic profiling (FoundationOne® assay) (Fig. 4). 23 genes were identified with single nucleotide variants (SNVs), 7 of which were truncal and present in all tumor samples obtained pre- and post-atezolizumab (−556d, −231d, +380d, +689d). Truncal copy number alterations (CNAs) of ZNF703, FGFR1, MYST3 and GPR124 were detected in all samples, while frequencies of CNAs in PIK3CA, IRS2, MYC, FAT1, CUL4A, MYC and CEBPA were less consistent but were present in all samples below the assay’s validated reporting threshold. Deleterious mutations in the tumor suppressors TP53 and RB1 emerged as possible oncogenic driver mutations, with amplification of the epigenetic activator MYST3 as a possible modifier. Subclonal somatic SNV mutations peaked post-XRT and were not detected after atezolizumab exposure. This observation is consistent with the possibility that radiation generated neoantigens targeted by anti-tumor T cells, further activated by atezolizumab. Similarly, the TMB was highest post-XRT (TMB = 8.11 Mut/Mb, −231d), and lowest post-atezolizumab (TMB = 2.7 Mut/Mb, +380d, reference in Additional file 1: Table S1). The temporal evolution of the tumor genomic landscape (SNV and TMB) suggests that low frequency clones appear during cancer therapies. It is possible that atezolizumab might have activated T-cells targeted against immunogenic tumor cell clones.Fig4Fig. 4

Characterization of Genomic Landscape Over Time. Samples collected pre- and post-atezolizumab exposure were tested with the FoundationOne® targeted NGS assay. Upper panel: genes with detected single nucleotide variants (SNV). Mutant allele frequencies (MAF) are shown for each specimen. Asterisk (*) indicates that the variant was present at a frequency below the validated reporting threshold. Light gray: predicted somatic mutations, dark gray: predicted germline mutations; Bold: predicted subclonal somatic mutations. Middle panel: genes with detected copy number alterations (CNAs). Numbers indicate the number of copies detected. Asterisk (*) indicates that low-level amplifications were detected below the validated reporting threshold of > 5 copies. No homozygous deletions were observed. Lower panel: tumor mutational burden (TMB) indicated as mutations per megabase