Spatial biology captures the effects of neoadjuvant chemo-immunotherapy in lung cancer

Abstract

PD-1 pathway blockade in combination with chemotherapy has emerged as a treatment paradigm for patients with resectable lung cancer, but insights into predictive biomarkers and mechanisms of immune responses are lacking. A study uses spatial transcriptomic methods to identify patterns within the tumor microenvironment associated with response.

Over the past decade, blockade of the PD-1 signaling pathway has revolutionized treatment of a growing number of advanced malignancies. Based on this success, recent efforts have centered on moving PD-1 and PD-L1 inhibitors into the management of patients with earlier stage cancers. Non-small cell lung cancer (NSCLC) has been at the forefront of this approach. Four recent large phase 3 studies have demonstrated considerable improvements in clinical outcomes with neoadjuvant (that is, before surgery) or perioperative (that is, before and after surgery) PD-1 pathway blockade and chemotherapy compared with chemotherapy alone in patients with resectable NSCLC, establishing this approach as a new standard¹. Despite these advances, a sizable proportion of patients do not respond to PD-1 or PD-L1 inhibitors, underscoring the need to identify new predictive biomarkers for these agents and to better define the underlying biology of anti-tumor immune responses. In this issue of Nature Genetics, Yan et al.² examine a cohort of pre- and post-treatment tumor samples from patients with NSCLC receiving neoadjuvant PD-1 inhibitors with chemotherapy, describing tumor cell states and the spatial organization of the tumor microenvironment (TME) of responders versus non-responders. Such studies extend our collective understanding of the determinants of response to PD-1 pathway blockade.

During the initial clinical development of PD-1 pathway inhibitors, early studies identified PD-L1 expression, high tumor mutation burden, microsatellite instability and enhanced CD8 T cell infiltration as validated predictive biomarkers of immune checkpoint inhibitor (ICI) response across multiple tumor types³. Translational studies have also nominated additional predictive biomarkers of ICI outcomes in NSCLC, including mutational signatures (for example, smoking signature), defects in antigen presentation (for example, human leukocyte antigen loss of heterozygosity), transcriptomic signatures (for example, T cell-inflamed gene expression profiles), single-gene mutations (for example, in EGFR, STK11 and KEAP1), and gut microbiota composition, among others^3,4. Of note, most biomarker studies to date have focused on analysis of pre-treatment biopsy specimens alone, as obtaining on-treatment and/or post-treatment samples has not been routinely feasible in advanced NSCLC outside of clinical trials. Crucially, with the introduction of neoadjuvant PD-1 pathway blockade as a new paradigm in resectable NSCLC, investigators now have a unique opportunity to routinely obtain and analyze paired, pre- and post-ICI specimens.

Contemporaneous with therapeutic advances in cancer immunotherapy, technological advances in transcriptomics are empowering researchers to build comprehensive atlases of human tumors. Single-cell RNA sequencing (scRNA-seq) and spatially resolved transcriptomic microscopy have enabled interrogation of human tumor composition and organization at unprecedented resolution. One such spatial technology is the 10x Genomics Visium platform, which uses spatially indexed arrayed bins of capture oligonucleotides to create a transcriptomic map of tissue. The bin boundaries do not align with cell boundaries and generally capture more than one cell at a time, but deconvolution methods can computationally infer the composition of cells within each bin. The NanoString GeoMx Digital Spatial Profiler platform is a different spatially indexed technology, somewhat analogous to laser capture microdissection, that allows identification of transcripts expressed within a user-defined region of interest. With these and other emerging spatial technologies, scientists now have powerful tools to analyze cell composition and transcriptional changes in the TME (Fig. 1) — tools that are well-suited to interrogate tumor cell-intrinsic and -extrinsic factors associated with immunotherapy responses in patients.

Fig. 1 |. The composition and organization of the TME reflect the interaction of numerous features.

The composition of the TME is influenced by: 1) tumor-intrinsic features, 2) quality of antigens, 3) host factors and 4) exposure to anti-cancer therapies. MHC, major histocompatibility complex; OPN, osteopontin.

Yan et al.² leverage the three above transcriptomic approaches to explore biomarkers of response to neoadjuvant chemotherapy-ICIs in 27 tumor samples from 19 patients with NSCLC. They examined pre- and post-therapy specimens by scRNA-seq. Of note, only seven patients had matched pre- and post-treatment tumor specimens for scRNA-seq, highlighting one challenge of deploying this technology to small biopsy or resection specimens in NSCLC. Pre-treatment specimens were spatially profiled by GeoMx and post-treatment specimens by Visium.

Using scRNA-seq analysis of pre-treatment carcinoma cells, the authors found that patients who subsequently developed a pathological response to chemotherapy-ICIs had higher interferon-stimulated gene (ISG) scores, including for the chemokines CXCL9, CXCL10 and CXCL11, than pathological non-responders. These findings join a growing body of literature across multiple cancer types showing that this group of genes is predictive of response to immunotherapy^5–9. Next, Visium was used to analyze post-treatment specimens, demonstrating that CXCL9 localized to the tumor–stromal interface and co-localized with cytotoxic T cells. These findings are consistent with a positive feedback loop in which T cell-expressed interferon-γ (IFNγ) induces neighboring tumor and myeloid cells to express ISGs, including CXCL9, CXCL10 and CXCL11 — chemokines that attract yet more activated T cells expressing IFNγ. This spatially organized multicellular network, or ‘immunity hub’, has been identified not only in human lung cancer, but also in melanoma^7,10 and triple-negative breast cancer¹¹.

Beyond ISG scores, Yan et al.² also found that patients who subsequently responded to therapy had lower NRF2 target scores in scRNA-seq analysis than non-pathological responders. NRF2 is a central transcription factor in the pathway that helps cells mitigate and survive oxidative stress¹². NRF2 signaling is frequently elevated in cancer, often through mutation of the negative regulator KEAP1 (ref. 12). KEAP1 mutations in KRAS-mutant NSCLC are associated with lower response rates to PD-1-blockade¹³. Interestingly, KEAP1-mutant lung cancers may have differential sensitivities to different types of ICIs, as the addition of CTLA4 inhibition to PD-(L)1-chemotherapy has been shown to resensitize these tumors to therapy¹⁴. More work is therefore needed to better define the clinicogenomic features of patients with high NRF2 target scores and examine the antitumor activity of combined PD-1 and CTLA4 inhibition — ultimately moving the field toward more precision immunotherapy.

To complement the above scRNA-seq analysis, Yan et al.² performed spatial transcriptomic analysis on resected NSCLC specimens, observing enrichment of COL11A1⁺GREM1⁺ cancer-associated fibroblasts (CAFs) and SPP1⁺ macrophages adjacent to carcinoma cells, and the frequency of these cells post-treatment correlated with tumor burden and lack of response to therapy. These findings are consistent with previous reports that SPP1⁺ macrophages negatively correlate with ICI response⁹. The authors speculate that the COL11A1⁺ CAFs and SPP1⁺ macrophages may collaborate with carcinoma cells to restrain anti-tumor immunity, possibly by physically obstructing access of anti-tumor T cells to carcinoma cells. Moving forward, functional studies will be needed to validate this mechanism of immune evasion and test whether this could inform new therapeutic approaches. Of note, in contrast to the enrichment of COL11A1⁺GREM1⁺ CAFs in tumor regions, areas without carcinoma cells tended to have ADH1B⁺ fibroblasts. Such analyses raise important questions, including how best to define the borders of tumors and regions of interest in ICI-treated tumors, and whether ADH1B⁺ fibroblasts should still be considered CAFs, as they were found outside of regions with tumor cells.

In prior studies, tertiary lymphoid structures (TLSs) have been associated with favorable response to immunotherapy¹⁵. Here, Yan et al.² developed a tool to identify TLSs as hematoxylin-dense areas after hematoxylin and eosin staining, with corresponding Visium areas showing both T cell and B cell signatures. Using unbiased clustering analysis of their Visium data, they defined four TLS states, which they postulate to be stages in a temporal progression: (1) ‘lymphoid aggregates’, characterized by IL7R expression; (2) ‘activated TLSs’, rich in germinal center (GC) B cell signatures; (3) ‘declining TLSs’, characterized by low GC and T cell activity; and (4) ‘late TLSs’, defined by T cell IFNG and HAVCR2 (encoding TIM3) exhaustion signatures.

The authors found that on-treatment resection specimens from complete responders overwhelmingly contained late TLSs, but these specimens lacked active TLSs. These data suggest that the anti-tumor cytolytic T cell response may be a more characteristic feature than the presence of GCs with respect to response to chemo-immunotherapy. However, it is noteworthy that active TLSs were not observed in the absence of tumor. Non-responders and the one patient with a major pathological response showed activated TLSs characterized by a GC signature. The authors also found an inverse correlation between hypoxia signature and activated TLSs. The meaning of this correlation remains to be determined, as does whether a GC response contributes to anti-tumor immunity.

Collectively, these data serve as an important resource for the field and an early model of how to conduct neoadjuvant translational studies. Moving forward, we anticipate a rapid increase in such translational studies due to a dramatic expansion in the number of neoadjuvant immunotherapy clinical trials that have been launched across various tumor types. Beyond affording more rapid read-outs of anti-tumor activity than adjuvant studies (typically 2–3 months versus years, respectively), neoadjuvant studies also serve as a platform for new drug development and translational insights into mechanisms of action. In parallel with these therapeutic advances, future directions include incorporation of emerging spatial technologies (for example, Xenium, MERFISH and CosMx) applied to both pre- and on-treatment specimens. Such work with true single-cell spatial resolution will further define the fibroblast and immune cell populations that drive anti-tumor immune responses in NSCLC. In parallel, larger cohorts will also be necessary to remove confounding variables, such as tumor histology. Ultimately, by mapping dynamic changes in the TME before and after immunotherapy, we will gain new insights into predictive biomarkers of response and, ideally, nominate new therapeutic combinations.