Document resource
WHAT IS ALREADY KNOWN ON THE TOPIC A series of studies have identified genetic and immune determinants of therapeutic response to immune checkpoint inhibitors (ICIs) in metastatic renal cell carcinoma (RCC).However, the determinants of response in the primary tumour remain largely undefined.WHAT THIS STUDY ADDS Through a genomic and transcriptomic analysis of pretreatment and post-treatment primary RCC tumours, this study identified gene expression programmes associated with therapeutic response and dynamic change in immune pathways in the primary tumour.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY This study thus has implications for understanding primary tumour ICI response and resistance in the metastatic context where the primary is still in place and in the neoadjuvant setting.Introduction Over the past decade, the first-line treatment of advanced renal cell carcinoma (RCC) has shifted from single-agent vascular endothelial growth factor (VEGF) targeted therapies to combination therapies that include immune checkpoint inhibitors (ICIs). 1–3 ICIs, including agents targeting the programmed death 1 (PD-1) axis, have delivered durable anti-tumour responses and acceptable safety profiles as first-line and second-line treatments in patients across multiple tumour types, including RCC.2 4–6 ICIs have transformed care of advanced RCC and are now standard first-line agents of care.1–3 7Gene expression and genomic studies from pivotal clinical trials have identified molecular biomarkers which influence patient response to ICIs in advanced RCC. The IMmotion150 and IMmotion151 trials showed that angiogenic signatures predict sensitivity to VEGF-targeted therapy, such as sunitinib, whereas tumours with high T-effector signatures forecast ICI benefit.8 9 Analysis of the JAVELIN Renal 101 trial extended this observation, linking a T-effector signature to avelumab+axitinib response.10 Additional studies have connected T-cell clonality, clonal neoantigen load, tertiary lymphoid structure (TLS)-enriched microenvironments and loss-of-function PBRM1 mutations to superior ICI outcomes.5 11–14While previous research has significantly advanced our understanding of the systemic response to ICIs in advanced RCC, the clinical trials responsible for ICI approvals primarily included patients who had already undergone nephrectomy.1 2 7 15 However, in current clinical practice, many patients with metastatic RCC still have their primary tumour in place. Consequently, the molecular features that govern therapeutic efficacy specifically within the primary tumour remain unclear yet are critically important to understand. Further, most studies have focused only on the analysis of pretreatment tumour samples, which fail to capture dynamic changes in the tumour microenvironment that occur with ICI therapy. To address these gaps in knowledge, we conducted a comprehensive genomic and transcriptomic analysis of paired pretreatment and post-treatment primary RCC tumours treated with ICI, aiming to define the tumour-intrinsic and microenvironmental factors that shape ICI response.Materials and methods Specimen processing Molecular profiling was performed at Caris Life Sciences (Phoenix, Arizona, USA), a College of American Pathologists (CAP)/Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory. H&E-stained formalin-fixed, paraffin-embedded (FFPE) slides of the patient’s tumour underwent review by a board-certified pathologist or trained pathologist assistant. Tumour enrichment was achieved by harvesting targeted tissue using manual microdissection techniques. For MI Tumor Seek Hybrid, a minimum of 20% tumour nuclei in the area for microdissection was required, with a minimum 10 mm 2 dissection area.MI Tumor Seek Hybrid/MI Cancer Seek Total nucleic acid was auto extracted using a MagMax FFPE DNA/RNA Ultra extraction kit (ThermoFisher Scientific, Waltham, Massachusetts, USA). To perform simultaneous DNA and RNA sequencing using the MI Tumor Seek Hybrid assay, library preparation was performed on the Bravo Automated Liquid Handling Platform (Agilent, Santa Clara, California, USA) using KAPA Library Prep reagents (Roche, Indianapolis, Indiana, USA) and custom cDNA primers (IDT, Coralville, Iowa, USA; GeneLink, Elmsford, New York, USA). RNA was labelled during first-strand cDNA synthesis. Custom KAPA baits panels were designed to enrich for 720 clinically relevant genes at high coverage and high read-depth and an additional >20 000 genes at lower depth, along with single nucleotide polymorphism (SNP) and pathogen panels (Roche). Sequencing was performed on the NovaSeq 6000 System (Illumina, San Diego, California, USA) (RRID:SCR_016387) using recommended reagents. The average sequencing depth of this assay is 230x for 20 859 genes (whole exome), 1000x for 720 genes with known and potential clinical relevance, and 1500 x for 230 reportable genes. RNA was sequenced to a minimum of 1.37 million total mapped reads. Sequencing data were extracted into split FASTQ files (RNA and DNA) for further processing using Caris’s bioinformatics pipeline. For RNA, Transcripts Alignment to a Reference STAR (RRID:SCR_004463) software was used for alignment using the RNA FASTQ files. 16 Transcripts per million (TPM) molecules were generated using the Salmon expression pipeline (RRID:SCR_017036).17Variant calling Variants detected were mapped to reference genome (hg19/38) using the Burrows-Wheeler Aligner (BWA 0.7.17; RRID:SCR_010910) embedded in the analysis tools licensed from Sentieon. Bioinformatic tools including SamTools (RRID:SCR_002105), Pindel (RRID:SCR_000560) and snpEff (RRID:SCR_005191) were incorporated to perform variant calling functions and annotations were standardised to the Human Genome Variation Society format. Germline variants were filtered with various germline databases, such as 1000 Genomes (RRID:SCR_006828) and dbSNP (RRID:SCR_002338). All variants were detected with >99% confidence, with a reporting threshold of 5% variant allele frequency. Genetic variants identified were interpreted by board-certified molecular geneticists and categorised as ‘pathogenic,’ ‘likely pathogenic,’ ‘variant of unknown significance,’ ‘likely benign’ or ‘benign,’ according to the American College of Medical Genetics and Genomics standards. Pathogenic and likely pathogenic variants were counted as ‘reportable.’Copy number variations (CNVs) In the MI Tumor Seek Hybrid assay, CNVs were determined using CNVkit. 18 Amplification status was determined for 325 clinically relevant genes and deletion status for 220 clinically relevant genes. A gene was determined to be amplified if segment-level copies were ≥6 with the 95% prediction interval >3.5. A gene was considered deleted if segment-level copies <1 with the 95% CI <1 copy. 1p19q co-deletion was determined when both chromosome arm copy ratios <0.8 and all copies per arm were <2. Co-occurrence of chromosome 7+/10- was determined when chr7 >2.8 copies and chr10 <1.2 copies.Tumour mutational burden (TMB) TMB was measured by counting missense, nonsense, in-frame INDEL and frameshift variants found per tumour in the coding regions of genes analysed (25 Mb for Hybrid). Filtering was performed to remove low-quality and low-depth variants or variants determined to be unreliable or unassociated with TMB. Presumed germline variants found in databases such as dbSNP151 (RRID:SCR_002338) and Genome Aggregation Database (gnomAD) (RRID:SCR_014964) and found in at least 10% of training samples were also filtered. A cut-off point of ≥10 mutations per megabase (Mb) was used based on the KEYNOTE-158 pembrolizumab trial, 19 which showed that patients with a TMB of ≥10 mt/Mb across several tumour types had higher response rates than patients with a TMB of <10 mt/Mb. Samples with low depth of coverage (<100x) were considered indeterminate. Caris Life Sciences is a participant in the Friends of Cancer Research TMB Harmonization Project.20Loss of heterozygosity (LOH) The 22 autosomal chromosomes were split into 552 segments (2–6 Mb in size) and the LOH of single nucleotide polymorphisms (SNPs) within each segment was calculated. A t-test was performed on regions for the presence of LoH with a minimum of 0.15 Mb required as the smallest LoH segment. Caris whole-exome sequencing (WES) data consist of approximately 250K SNPs spread across the genome (17 SNPs/Mb), with 200K from exonic regions and 50K from intronic regions (Agilent). The MI Tumor Seek Hybrid assays use a KAPA HyperChoice MAX 5 Mb Baits SNP panel (Roche). The final call of genomic LOH is based on the percentage of all 552 segments with observed LOH (high ≥16%, low <16%; if fewer than 3000 SNPs can be read, the test is reported as indeterminate). Segments excluded from the calculation of genomic LOH include those spanning ≥90% of a whole chromosome or chromosome arm and segments which are not covered by the SNP backbone and the WES panel. A healthy normal tissue sample (NA12878) from epithelial cells was used as the normal non-LOH negative control to estimate noise. LOH was indeterminate if depth <200x.Microsatellite instability (MSI) status The threshold to determine MSI-high status from NGS data was 39 or more loci with frameshift mutations out of 5721 loci examined. These 5721 loci were selected based on sequencing coverage and enrichment of frameshift mutations in MSI-high patient samples. The sample was indeterminate if sequencing depth of a panel of reference genes (n=753) was <100x.HLA genotyping (WES/Hybrid pipeline) The Caris WES HLA assay provides a genotype for the MHC Class I HLAs (A, B and C genes) down to the allele using OptiType, 21 when reads are above 10x depth.Difference gene expression and gene set enrichment analysis (GSEA) Differential gene expression analysis was performed using raw RNA counts with DESeq2. 22 All genes were used for GSEA, conducted using fgsea.23 Parameters for GSEA were: minSize=15, maxSize=500 and nPermSimple=20 000 and used the hallmark gene sets from MSigDB.24Gene expression signature scoring Gene expression signature scoring was performed using transcript per million-normalised RNA data. TPM-normalised values were log-normalised then ssGSEA was performed for each gene list using corto. 25 Gene set for exhausted T cells included BTTLA, CTLA4, HAVCR2, LAG3, PDCD1 and TIGIT.Statistical analysis All statistical tests were performed in R. The function wilcox.test was used to compare distributions. All analyses were two-sided, and p values were Benjamini-Hochberg corrected. 26Results Cohort details A cohort of tumour specimens comprising (1) pretreatment biopsy samples and (2) post-treatment nephrectomy samples was assembled. Diagnostic biopsy samples were not always available for research use. The total cohort included 46 tissue samples consisting of 15 biopsy and 31 nephrectomy samples. All samples underwent whole-exome and bulk RNA sequencing ( figure 1a), and clinical response was determined by radiographic imaging and tissue pathology, with response defined as >30% radiographic shrinkage (investigator assessed based on measurements of the primary tumour). Imaging was performed at regular intervals at the discretion of the treating physician, in alignment with standard of care.Figure 1Analysis of 46 renal cell carcinoma (RCC) samples stratified by treatment status and response. (a) Schematic representation of the study cohort. Biopsy (n=15) and nephrectomy (n=31) samples were collected, with 13 patients having paired samples. All biopsy and nephrectomy samples underwent bulk RNA sequencing (RNA-Seq) and whole-exome sequencing, while clinical response was determined by radiographic imaging. (b) Patient breakdown summarising sample collection (nephr, nephrectomy), tumour histology (non-CC, non-clear cell), therapy received (Tx class; ICI, immune checkpoint inhibition; TKI, tyrosine kinase inhibitor) and clinical response (NR, non-responder; R, responder). (c) Alluvial plot depicting tumour stage from biopsy to nephrectomy. ICI therapy was associated with tumour shrinkage in some primary RCC tumours, as evidenced by changes in tumour stage and radiographic response.A total of 33 patients were sampled, where 13 had paired samples and 20 patients had only one of two tissue timepoints. 17 samples were from responders (n=4 biopsy, n=13 nephrectomy) and 29 samples were from non-responders (<30% shrinking, investigator assessed), n=11 biopsy, n=18 nephrectomy). Among the 33 patients, 26 were diagnosed with clear cell RCC, 2 were diagnosed with clear cell papillary RCC and 5 patients had unclassified RCC (figure 1b). Most patients (n=19) were treated with nivolumab+ipilimumab, while others were treated with only ICI (nivolumab, n=3), an ICI plus an anti-angiogenic tyrosine kinase inhibitor (TKI; lenvatinib+pembrolizumab, n=2; pembrolizumab+axitinib, n=4) or the TKI sunitinib alone (n=2) immediately prior to surgery. The remaining three patients received other therapies (figure 1b; online supplemental table 1).SP210.1136/bmjonc-2025-001031.supp2Supplementary dataFor patients with nephrectomy tissue available (and therefore, pathologic staging information), we found responder tumours (n=13) were either stable or downstaged at the time of nephrectomy (compared with pretreatment clinical staging) (figure 1c; online supplemental table 1). As expected, non-responder tumours (n=20) generally remained stable or pathologically upstaged from the initial clinical staging, apart from 2 patients whose tumours decreased in stage (from stage 4 to 3 and stage 2 to 1), likely due to differences between clinical and pathologic staging.Identification of somatic mutations in RCC across ICI evolution WES was performed on all samples, identifying an average of 17.6 mutations per sample, with individual sample mutation counts ranging from 6 to 95. Highly mutated genes included canonical RCC driver alterations in VHL (70%), PBRM1 (20%), SETD2 (11%), PTEN (13%), BAP1 (17%) and KDM5C (13%) ( figure 2a; online supplemental table 2). Of the 10 patients with a VHL mutation in their biopsy, 3 (30%) did not have a VHL mutation in their post-treatment nephrectomy, which may be a technical artefact due to regional heterogeneity or technical issues. There were no significant associations between the prevalence of recurrent driver mutations (present in three or more samples) and ICI response in pretreatment or post-treatment tumours (p>0.05 for all comparisons; Fisher’s exact test).Figure 2Genomic landscape is stable across treatment and response groups. (a) Oncoprint showing genomic alterations across samples, grouped by patient pairs, sample type and treatment response. (b) Tumour mutation burden (TMB) stratified by sample type and response (NR, non-responder; R, responder). Biopsy samples from responders exhibit significantly higher TMB compared with their nephrectomy samples (uncorrected Wilcoxon test p value=0.04). (c) Variant allele frequencies (VAF), normalised by tumour purity, plotted before and after treatment.Next, we focused on three genomic features with mechanistic links to anti-tumour immunity. TMB has seen efficacy serving as a biomarker for predicting ICI response across several cancer types,27 28 although the efficacy of TMB as a biomarker is low in metastatic RCC.14 We examined whether (1) TMB might be predictive of response in the primary tumour at baseline; and (2) whether TMB might decrease with treatment in ICI-responsive tumours as an indication of effective immunoediting and ‘pruning’ of certain tumour clones. At baseline, there was no difference in TMB between responder and non-responder tumours. Across our full cohort (including both paired and unpaired samples), TMB in biopsy samples from responders was significantly higher than from responder nephrectomy samples (figure 2b; Wilcoxon p value=0.04), suggesting a potential deletion of subclones as a result of immunoediting. However, it is possible that confounding factors, such as differences in tumour purity, sampling variation, or tumour heterogeneity, could impact this finding. To investigate whether ICI selectively prunes tumour clones, we investigated only samples that had paired biopsy and nephrectomy samples. We normalised variant allele frequencies (VAFs) of driver mutations mutated in 3+ samples both pretreatment and post-treatment by tumour purity, as estimated using the ESTIMATE package,29 30 and found no consistent differences in normalised VAF with treatment pretreatment and post-treatment (figure 2c; online supplemental table 2), suggesting ICI therapy does not broadly alter driver clonality. Thus, the observed decrease in TMB may reflect changes in immune infiltration or possibly immune pruning of smaller subclonal populations rather than large shifts in clonal architecture. We further investigated LOH, which can facilitate immune escape by deleting tumour-suppressor or antigen-presenting alleles,31 and HLA haplotype evolutionary divergence (HED), which captures germline diversity in peptide binding which may broaden neoantigen display.32 Neither LOH nor HLA HED were significantly associated with response (online supplemental figure 1; online supplemental table 2). Taken together, these findings suggest that while canonical features of anti-tumour immunity, such as LOH and HLA HED, are not predictive of response, subtle immunoediting events, including a decrease in TMB without shifts in clonal drivers, may contribute to tumour remodelling in tumours that respond to ICI.SP110.1136/bmjonc-2025-001031.supp1Supplementary dataElevated immune-related pathways in pretreatment ICI-responsive primary RCC tumours We next sought to understand the detailed molecular features of pretreatment primary RCC tumours that drive immunotherapy response and resistance. Through differential gene expression analysis, 14 genes were significantly elevated in responders compared with non-responders ( figure 3a; online supplemental table 3). Many of the 14 genes encoded were immunoglobulin components, including components of immunoglobulin heavy chains (IGHG4: false discovery rate (FDR)-adjusted p value=0.001, IGHJ3: FDR-adjusted p value=0.001, IGHV6-1: FDR-adjusted p value=0.013, IGHV3-15: FDR adjusted p value=0.037) and immunoglobulin light chain components (IGLC1: FDR-adjusted p value=0.049, IGLC7: FDR-adjusted p value=0.049). IGLL5, part of pre-B cell receptors,33 was also significantly elevated (FDR-adjusted p value=0.049).Figure 3Immune-related gene expression signatures are associated with immunotherapy response in renal cell carcinoma (RCC) biopsies. (a) Volcano plot of genes differentially expressed between biopsies from responder (R) versus non-responder (NR) patient tumour samples coloured by significance. Genes in red are significant (false discovery rate (FDR)-adjusted p value <0.05 and absolute log2 fold change >0.5). (b) Gene set enrichment analysis results comparing biopsies from responder versus non-responder patients. All pathways shown are statistically significant (FDR-adjusted p value <0.05). (c) Boxplots comparing gene signature expression in responder versus non-responder biopsy samples (uncorrected Wilcoxon test p value): B cells from Bindea et al34 (left), tertiary lymphoid structure (TLS) signature from Xu et al36 (middle) and TLS signature from Meylan et al35 (right). Elevated expression of B cell–related and TLS-related gene programmes at baseline is associated with immunotherapy response in RCC.We next assessed whether differences in cellular pathways may differ between responsive and non-responsive tumours through GSEA. At a pathway level, biopsies from responsive tumours showed robust elevation of immune-related signatures, including interferon gamma response (normalised enrichment score (NES) = 2.94, FDR-adjusted p value=2.24×10-25), interferon alpha response (NES=2.80, FDR-adjusted p value=1.19×10-14), inflammatory response (NES=2.13, FDR-adjusted p value=9.99×10-9), IL6-JAK-STAT3 signalling (NES=1.85, FDR-adjusted p value=4.02×10- 4) and IL2-STAT5 signalling (NES=1.59, FDR-adjusted p value=2.33×10-3) (figure 3b; online supplemental table 3). In contrast, biopsies from non-responding tumours were enriched for fatty acid metabolism (NES=1.49, FDR-adjusted p value=1.44×10-2) and oxidative phosphorylation (NES=2.05, FDR-adjusted p value=7.37×10-8). These findings indicate that, prior to treatment, tumours primed for response to ICIs are characterised by enrichment of immune pathways and reduced expression of metabolic programmes, including fatty acid metabolism and oxidative phosphorylation.To further characterise the immune differences between responder versus non-responder pretreatment tumour samples, we compared the expression of previously reported transcriptomic immune signatures (figure 3c; online supplemental table 3). Responder tumours exhibited elevated expression of a B cell signature (from Bindea et al34; uncorrected Wilcoxon p value=0.04). We also investigated two separate signatures of TLS, one from Meylan et al35 and Xu et al,36 where the Xu et al signature was statistically significant (uncorrected Wilcoxon p value=0.04) and the Meylan et al signature narrowly missed significance (uncorrected Wilcoxon p value=0.056). Gene expression signatures associated with T cell infiltration (JAVELIN101 Teff,10 McDermott Teff8), myeloid infiltration (McDermott Myeloid8) and angiogenesis (JAVELIN101 Angio10, McDermott Angio8) that were previously shown to predict therapeutic response in RCC did not show significant associations with response in this primary tumour cohort (online supplemental figure 2; online supplemental table 3). Together, these data suggest that, prior to treatment, ICI-responsive tumours are defined by an active immune microenvironment, including B cell–rich TLS niches.Characterisation of transcriptomic evolution on ICI therapy in ICI-responsive and ICI-resistant primary RCC tumours To understand how transcriptomic programmes evolve across the therapeutic axis, we leveraged our 13 paired samples, comparing pretreatment biopsies to post-treatment nephrectomies from our 4 ICI responding and 9 non-responding patients. In ICI responding patients, 2284 genes were significantly differentially expressed between biopsy and nephrectomy samples ( figure 4a; online supplemental table 4). Among the most significantly upregulated genes in post-treatment samples were members of the keratin-associated protein (KRTAP) family (KRTAP9-1: FDR-adjusted p value=9.76×10-7, KRTAP5-7: FDR-adjusted p value=1.28×10-6, KRTAP4-1: FDR-adjusted p value=9.05×10-6, KRTAP5-3: FDR-adjusted p value=1.04×10-5). KRTAP genes, involved in epidermal differentiation, may reflect tissue remodelling and a mesenchymal to epithelial transition.Figure 4Divergent transcriptomic evolution during treatment in responders versus non-responders. (a) Volcano plot of genes differentially expressed between responder biopsy (Bx) versus nephrectomy (Nx) samples coloured by significance. Genes in red are significant (false discovery rate (FDR)-adjusted p value <0.05 and absolute log2 fold change >0.5). (b) Gene set enrichment analysis (GSEA) results comparing biopsies versus nephrectomies from paired samples from responder (R) patients. All pathways shown are statistically significant (FDR-adjusted p value <0.05). (c) Volcano plot of genes differentially expressed between non-responder biopsy versus nephrectomy samples coloured by significance. (d) GSEA results comparing biopsies vs nephrectomies from paired samples from non-responder (NR) patients. All pathways shown are statistically significant (FDR-adjusted p value <0.05). (e) Boxplots showing ssGSEA scores for TNF-alpha signalling (left), IL6-JAK-STAT3 signalling (middle) and inflammatory response (right) pathways in paired biopsy and nephrectomy samples coloured by response category. FDR-corrected GSEA p values. (f) Boxplots showing gene signature expression for dendritic cell, mast cell, exhausted T cell (TEx), central memory T cell (Tcm), Hallmark TLS from Cabrita et al37 and TLS from Xu et al36 signatures in paired biopsy (Bx) and nephrectomy (Nx) samples coloured by response category. Uncorrected Wilcoxon test p values. Immune-related transcriptional programmes demonstrate opposing trajectories of immune activation during treatment.At the pathway level, we found most significantly altered pathways were elevated in pretreatment samples (figure 4b; online supplemental table 4). Multiple immune pathways including interferon gamma response (NES=3.15, FDR-adjusted p value=1.74×10-31), interferon alpha response (NES=3.08, FDR-adjusted p value=2.10×10- 20), IL6 JAK STAT3 signalling (NES=1.90, FDR-adjusted p value=7.55×10-5) and IL2 STAT5 signalling (NES=1.85, FDR-adjusted p value=1.36×10-7) were all enriched before treatment. Pathways associated with metabolism such as oxidative phosphorylation (NES=3.34, FDR-adjusted p value=2.17×10-36) and glycolysis (NES=2.31, FDR-adjusted p value=1.01×10-12) were also elevated in pretreatment. None of the evaluated transcriptional signatures were significantly associated with sample type (ie, pretreatment or post-treatment) in responding tumours (online supplemental figure 3; online supplemental table 4).Moving our focus to non-responder samples, we found 1511 genes were significantly differentially expressed between pretreatment and post-treatment primary RCC tumours (figure 4c; online supplemental table 4). Of these 1511 genes, several were associated with cellular stress and T cell exhaustion (FOS: FDR-adjusted p value=5.96×10-13, FOSB: FDR-adjusted p value=1.28×10-11, EGR1: FDR-adjusted p value=2.53×10-5, ATF3: FDR-adjusted p value=2.74×10-5, NR4A1: FDR-adjusted p value=3.77×10-5), all of which were elevated in post-treatment samples. CXCR4, a GPCR which acts as a chemokine receptor, was also significantly elevated after treatment (FDR-adjusted p value=3.87×10-6).Further, pathway analysis revealed a significant elevation in immune pathways in post-treatment nephrectomy samples only in the non-responding tumours, including TNF alpha signalling response (NES=2.15, FDR-adjusted p value=5.28×10-11), IL6 JAK STAT3 signalling (NES=1.50, FDR-adjusted p value=2.56×10-2) and inflammatory response (NES=1.79, FDR-adjusted p value=8.98×10-6) (figure 4d; online supplemental table 4).To directly compare response dynamics across patients, we performed single sample gene set enrichment analysis (ssGSEA) on pathways that were differentially expressed in both groups (TNF alpha signalling, IL6 JAK STAT3 signalling, and inflammatory response). In ICI-responsive tumours, activity of these pathways decreased post-treatment (FDR-adjusted p value=1.02×10-7, 7.55×10–5 and 3.61×10–3, respectively); conversely, they increased in non-responsive tumours (FDR-adjusted p value=5.28×10-11, 2.56×10–2 and 8.98×10–6, respectively) (figure 4e; online supplemental table 4), further suggesting divergent transcriptomic remodelling trajectories in ICI response and resistance.Comparing transcriptional immune signatures, we found that signatures for mast,34 dendritic34 and exhausted T cells, as well as two TLS signatures,36 37 were significantly elevated in post-treatment tumours only in non-responders (unadjusted p values: mast cells (responders=0.2, non-responders=0.024), dendritic cells (responders=0.057, non-responders=0.014), exhausted T cells (responders=0.89, non-responders=0.02), Cabrita et al Hallmark TLS (responders=0.69, non-responders=0.019), Xu et al TLS (responders=0.11, non-responders=0.024)). In opposition, a signature for T central memory (Tcm) cells was significantly depleted only in non-responders (responders=0.34, non-responders=0.03) (figure 4f). No other signatures were significant between pretreatment and post-ICI treatment primary tumours in non-responders (online supplemental figure 3; online supplemental table 4). While these analyses capture the inferred differences of these populations at two timepoints, they do not capture the true dynamics. It is possible that the dynamic changes, for example, in T cell exhaustion, may be clinically important. Nevertheless, these findings reveal that ICI induces distinct immune remodelling in responders and non-responders. Although non-responders did have an increase in signatures of TLS formation with immunotherapy, this was accompanied by a counterbalancing increase in exhausted T cells.Discussion ICIs have redefined systemic therapy for advanced RCC, yet recent clinical trials have enrolled patients after cytoreductive nephrectomy. Motivated by the numerous studies which have shown molecular signatures being predictive of therapeutic response in the metastatic setting, and the rise in cases where neoadjuvant therapy is provided with the primary tumour still in place, we investigated how the primary tumour changes with ICI treatment, focusing on the underlying genetic, transcriptomic and microenvironmental factors driving these changes.Previous studies by Braun et al14 and others38 39 showed no significant differences in ICI response based on CD8 T-cell abundance, suggesting that cell phenotype and organisation must also be considered. Consistent with this, we found ICI-responsive tumours showed enrichment of immune-related gene expression, including B cell and TLS signatures, as well as elevated interferon and inflammatory signalling. Recent work has supported the role of TLS in RCC: Meylan et al found mature TLS are correlated with response and progression-free survival,35 results which were validated using biospecimens from patients enrolled in a first-line phase II trial of the anti-PD-1 antibody nivolumab.13 Together, these data suggest then an immunologically enriched, TLS-rich microenvironment may predispose tumours to ICI responsiveness. These findings also raise the possibility that changes in TLS abundance during ICI treatment, rather than baseline levels alone, may be more informative for understanding treatment response.In contrast, non-responder tumours showed baseline enrichment for oxidative phosphorylation and fatty acid metabolism. Najjar et al found high tumour oxygen consumption limits anti-PD-1 efficacy driving T-cell exhaustion in murine melanoma tumours whereas melanoma cell lines lacking oxidative metabolism were responsive to anti-PD-1 therapy.40 Additionally, fatty acid uptake facilitated by CD36 induces ferroptosis and impairs antitumour ability of CD8+ T cells.41 Collectively, these results support our findings that oxidative phosphorylation and fatty acid metabolism lead to ICI-resistant tumours.Longitudinal profiling of paired tumours further revealed divergent transcriptional trajectories. Responders displayed a dampening of immune and metabolic pathways post-treatment, whereas non-responders demonstrated increased immune signalling and expression of exhaustion-related genes such as FOS, EGR1 and NR4A1.42 A prior study exploring tumour and immune adaptation during ICI therapy in patients with advanced RCC reported opposing findings where responder tumours showed increased infiltration of CD8+ T cells and reduced expression of immunosuppressive genes. However, their non-responder tumours likewise showed increased immune infiltration after ICI therapy and elevated expression of immune checkpoint genes, such as LAG3.43Our study has several notable limitations. The relatively small number of patients with paired samples constrains statistical power and limits generalisability, especially when evaluating longitudinal changes. Within clear cell RCC, there can be notable genetic and transcriptional heterogeneity, and so differences in sampling and the comparison of a small sampling (biopsy) with larger sampling (nephrectomy) specimen has inherent limitations. Further, although our cohort largely consists of clear cell histology, there are a number of patients included with non-clear cell (or ‘variant’) histology. While we believe it is important to investigate the biology of these variant tumours, as they represent an area of clinical need, they may contribute to heterogeneity. A re-analysis of our cohort with only clear cell histology yielded consistent trends, though statistical significance was lost due to smaller sample size (data not shown). The evaluation of treatment effects is also confounded by heterogeneity in treatments (eg, ICI, TKI), including different ICI agents (eg, pembrolizumab, nivolumab) and combinations (dual ICI vs ICI plus TKI). Anti-angiogenic agents have a substantially different impact on gene expression profiles compared with checkpoint inhibition, and particularly in a small cohort, this represents a substantial limitation. Although transcriptional immune signatures have been shown to correlate with immunofluorescence in prior RCC studies, validation with orthogonal approaches such as single-cell RNA-sequencing or spatial transcriptomics would strengthen the interpretation of cell-type–specific findings. Furthermore, all observations are correlative, and functional studies will be required to directly test the mechanisms inferred from transcriptomic changes.This study thus has implications for understanding primary tumour ICI response and resistance in the metastatic context where the primary is still in place, and in the neoadjuvant setting. One could envision prospective neoadjuvant clinical trials that select for patients with high TLS signatures in the baseline tumour, as a means to enrich for those most likely to benefit. Further, through the analysis of paired pretreatment and post-treatment tumour tissue, our study facilitates the understanding of dynamic tumour and immune changes that occur during ICI treatment. Together, these findings suggest that both baseline immune landscape and treatment-induced transcriptional remodelling shape the response to ICI in primary RCC, and that the primary tumour provides a valuable window into these dynamics. Given the limited sample size, these findings must be interpreted cautiously, and larger scale validation is needed.