Integrative molecular characterization of resistance to neoadjuvant chemoradiation in rectal cancer

Abstract

Purpose:

Molecular properties associated with complete response or acquired resistance to concurrent chemotherapy and radiation therapy (CRT) are incompletely characterized.

Experimental Design:

We performed integrated whole exome/transcriptome sequencing and immune infiltrate analysis on rectal adenocarcinoma tumors prior to neoadjuvant CRT (pre-CRT) and at time of resection (post-CRT) in 17 patients (8 complete/partial responders [R], 9 nonresponders [NR]).

Results:

CRT was not associated with increased tumor mutational burden or neoantigen load and did not alter the distribution of established somatic tumor mutations in rectal cancer. Concurrent KRAS/TP53 mutations (KP) associated with NR tumors and were enriched for an epithelial-mesenchymal transition transcriptional program. Furthermore, NR was associated with reduced CD4/CD8 T-cell infiltrates and a post-CRT M2 macrophage phenotype. Absent any local tumor recurrences, KP/NR status predicted worse progression-free survival, suggesting that local immune escape during or after CRT with specific genomic features contributes to distant progression.

Conclusions:

Overall, while CRT did not impact genomic profiles, CRT impacted the tumor immune microenvironment, particularly in resistant cases.

INTRODUCTION

Radiation therapy is used in the management of nearly two-thirds of cancers (1), often fulfilling the role of a curative treatment modality in place of surgery. Therapeutic radiation can be adapted to target tumors in various anatomical locations as well as various malignancies. The radiation dose and fractionation can be altered to maximize tumor-killing while sparing normal tissues (2). Radiation therapy is typically combined with concurrent chemotherapy (CRT) in locally advanced disease. When used neoadjuvantly, pathological downstaging is a surrogate of long-term outcome in many disease sites (3–7). For example, in rectal cancer (RC), approximately 9-20% of patients with locally advanced disease have a pathological complete response (pCR) to neoadjuvant CRT (8) while 20-40% of patients have little to no response (9,10). Predictive biomarkers of pCR remain to be established.

The major mechanism of radiation-induced cell killing is likely through DNA damage. There is, however, emerging evidence that radiation also has effects on the tumor microenvironment with variation based on anatomic site, tumor histology, and multiple other characteristics (11). These cell killing effects can be further augmented by combining radiation with radiosensitizing systemic agents (12). In addition, there is recent interest in the utility of radiation to alter the adaptive immune response to improve treatment outcomes by creating a local anti-tumor immune response that may be modulated into a systemic anti-tumor immune response with the use of immunomodulatory agents (13–15). Proposed mechanisms include possible creation of increased neoantigens or tumor mutational burden (TMB) through the DNA-damaging effects of radiation (16,17), the latter of which has been previously demonstrated to correlate with response after treatment with immune checkpoint inhibitors (18,19). Despite the widespread use of radiation therapy for solid tumors, there has been slow progress in predicting treatment outcomes to radiation to allow for personalization of therapy on an individual level (12,15). Biomarkers have been in use and ultimately transformed the field of systemic therapy while few predictive biomarkers are available for radiotherapy (12).

Using RC as our model (3,20), we hypothesized that a comprehensive assessment of patient-matched pre- and post-CRT specimens, examining both tumor-intrinsic and microenvironmental features from the tumor site, may reveal features associated with treatment response at the molecular level. To that end, we leveraged a cohort of locally advanced RC patients who underwent fluoropyrimidine-based CRT to a dose of 50.4 Gy followed by surgical resection and analyzed genomic tumor changes in the matched pre- and post- treatment rectal tumor samples to identify drivers of resistance to neoadjuvant CRT and thereby identify biomarkers for patient stratification.

METHODS

Patient population and samples

We retrospectively identified patients with biopsy-proven locally advanced rectal cancer (defined as T3-4 or N+) who received neoadjuvant fluoropyrimidine-based chemotherapy concurrently with 50.4 Gy radiation therapy, followed by surgical resection within 8-11 weeks between 2010 and 2016 (20). Patients then went on to receive adjuvant systemic therapy, which consisted of FOLFOX (21). Patients had to have documented written consent through the institutional review board-approved protocol that collects tissue and whole blood specimens on patients with gastrointestinal malignancies in accordance with the Declaration of Helsinki and all applicable legal regulatory requirements. There were 77 patients who met initial criteria. Eligible patients had to have sufficient tumor tissue in study specimens of formalin-fixed, paraffin-embedded (FFPE) tissue sections from surgical samples, as well as a germline DNA specimen that was extracted from either peripheral mononuclear cells or histologically normal rectal tissue. Twenty patients were identified with sufficient tissue available. All patients were arbitrarily identified with no prior knowledge of genomic tumor status. All samples had to pass standard quality control measures. We identified 34 pre- and post-CRT matched tumor samples from 17 patients in our final cohort. Nine and 8 patients were classified as nonresponders (no evidence of any pathologic downstaging, NR) and responders (pathologic complete response or pathologic partial response, R) respectively at surgery based on pathologic evaluation.

DNA extraction and whole exome sequencing

DNA extraction, whole exome library prep and sequencing was performed for the samples as previously described (22,23). Slides were cut from FFPE blocks and examined by a board-certified pathologist to select high-density cancer blocks and ensure high purity of cancer DNA. Biopsy cores were taken from the corresponding tissue block for DNA extraction. DNA was extracted using Qiagen’s QIAamp DNA FFPE Tissue Kit Quantitation Reagent (Invitrogen). DNA was stored at −20 °C.

Whole exome capture libraries were constructed from 100 ng of DNA from tumor and normal tissue after sample shearing, end repair, and phosphorylation and ligation to barcoded sequencing adaptors. Ligated DNA was size selected for lengths between 200 and 350 bp and subjected to exonic hybrid capture using The Broad Institute Genomics Platform Custom Illumina bait. The Illumina exome specifically targets approximately 37.7Mb of mainly exonic territory made up of all targets from the Agilent exome design (Agilent SureSelect All Exon V2), all coding regions of Gencode V11 genes, and all coding regions of RefSeq gene and KnownGene tracks from the UCSC genome browser (http://genome.ucsc.edu). The sample was multiplexed and sequenced using Illumina HiSeq technology.

Sequencing was performed to an average depth of 150 X. Data were analyzed using the Broad Picard Pipeline which includes de-multiplexing and data aggregation.

Quality control, variant calling

Initial data processing and analysis of exome sequence data were performed used Broad Institute pipelines and as previously described (23). Using the Broad Picard Pipeline for alignment, BAM files were uploaded into the Firehose infrastructure to manage intermediate analysis files executed by analysis pipelines. Quality-control modules in Firehose (24) were run to compare the tumor and normal genotypes and ensure concordance between samples. Of samples from 20 initial patients, 6 samples from three patients were abandoned because of high estimates of tumor contamination (25), inadequate coverage (<40x tumor average coverage), or low tumor purity (26). This yielded a final number of 17 total pairs of pre and post treatment tumors for analysis.

The MuTect algorithm (27) was applied to identify somatic single-nucleotide variants in targeted exons. Strelka (28) was used to identify small deletions or insertions, and alterations were annotated with Oncotator (29). Mutations were examined for distribution and type and confirmed using the integrative genomics viewer (30,31).

Transcriptome Capture Method cDNA Library Construction

Using established protocols (32), total RNA was assessed for quality using the Caliper LabChip GX2. The percentage of fragments with a size greater than 200nt (DV200) was calculated using software. An aliquot of 200ng of RNA was used as the input for first strand cDNA synthesis using Illumina’s TruSeq RNA Access Library Prep Kit. Synthesis of the second strand of cDNA was followed by indexed adapter ligation. Subsequent PCR amplification enriched for adapted fragments. The amplified libraries were quantified using an automated PicoGreen assay.

200ng of each cDNA library, not including controls, were combined into 4-plex pools. Capture probes that target the exome were added and hybridized for recommended time. Following hybridization, streptavidin magnetic beads were used to capture the library-bound probes from the previous step. Two wash steps effectively remove any non-specifically bound products. These same hybridization, capture and wash steps are repeated to assure high specificity. A second round of amplification enriches the captured libraries. After enrichment the libraries were quantified with qPCR using the KAPA Library Quantification Kit for Illumina Sequencing Platforms and then pooled equimolarly. The entire process is in 96-well format and all pipetting is done by either Agilent Bravo or Hamilton Starlet.

Pooled libraries were normalized to 2nM and denatured using 0.1 N NaOH prior to sequencing. Flowcell cluster amplification and sequencing were performed according to the manufacturer’s protocols using HiSeq 2500. Each run was a 76bp paired-end with an eight-base index barcode read. Data was analyzed using the Broad Picard Pipeline which includes de-multiplexing and data aggregation.

Neoantigen prediction

HLA-type was inferred using POLYSOLVER (33) which uses a normal tissue BAM file as input. It then employs a Bayesian classifier to determine the genotype for each patient. Neoantigens were predicted for each patient by defining all novel amino acid 9mers and 10mers resulting from mutations (23). We filtered out mutations with <3 supportive reads, or <30 total reads at the position. Neoantigen prediction continued based on whether predicted binding affinity to the patient’s germline HLA alleles was <500 nM using NetMHCpan (34). Correlations and associated p values between neoantigen load and R versus NR was performed using Mann-Whitney U tests, p-values of <0.05 were considered significant.

Purity/ploidy, clonal/subclonal mutational calls

Purity and ploidy for each sample was estimated using ABSOLUTE algorithm (35). This algorithm integrates variant allele frequency distributions and copy number variants to estimate absolute tumor purity and ploidy and infer cancer cell fraction (CCF), which is the proportion of cancer cells in the sample which contain each mutation. An ABSOLUTE extension algorithm (35) was used to construct an inferred phylogenetic tree with clones, subclones, and evolutionary relationships in pre and post treatment tumor samples. As described in Brastianos et al (36), clones and subclones were determined through Markov Chain Montecarlo sampling using Dirichlet process Mixture Models on pre- and post-CRT mutation CCFs, which assigns mutations to subclones without pre-specifying the number of subclones. Mutations inferred to be in a subclone with a CCF ≥ 0.8 were described as “clonal” while those inferred to be in a subclone with CCF < 0.8 were called “subclonal.” For each subclone, two CCFs were inferred; one CCF in the pre-treatment tumor and CCF in the post-treatment tumor (23).

Changes in mutational and neoantigen load

Changes in mutational, neoantigen, and indel load were calculated using a paired t-test of changes in paired samples with a null hypothesis of a difference of 0 (23). p<0.05 was considered to be statistically significant.

Discovery of resistance or sensitivity biomarkers

We used MutSig2CV (26) to identify significantly mutated genes across our cohort of pre-CRT and post-CRT tumors. Each altered gene in the pre-treatment tumors had a p-value calculated for mutational significance considering only mutations private to these samples. Similarly, a p-value of mutational significance considering only those mutations private to the post-treatment tumor was calculated. Adjustment for hypothesis testing was performed using a Benjamini-Hochberg FDR of 0.1 (23).

Gene expression profiling

Available RNA-Seq data were analyzed as previously described (37). Briefly, expression data were examined and adjusted for batch effects using ComBat (38) using the R Bioconductor package “sva” V3.8 (39). Gene set enrichment analysis (40) was run using https://genepattern.broadinstitute.org using 50 ‘Hallmark’ gene sets to investigate differences in gene set expression in R vs. NR (pre-CRT R vs. pre-CRT NR; post-CRT R vs. post-CRT NR) with 1000 permutations, type ‘gene_set.’ Gene level transcripts per million (TPM) were the input. Family-wise error rates were calculated to identify significant gene sets.

To determine the relationship between CRT and the immune landscape, we analyzed matched transcriptomes from the tumors using CIBERSORT (41) to deconvolute immune cell populations from bulk transcriptome data using immune-cell associated signatures. From this, we inferred overall immune infiltrate and relative immune cell populations in both the pre-CRT and post-CRT specimens. This was run using the CIBERSORT interface (https://cibersort.stanford.edu). The analysis was set to absolute quantification output. Input was gene level TPM and leukocyte gene signature matrix (LM22) (41) was used to deconvolve 22 immune cell subset populations. Absolute quantification normalizing by the 50th percentile of overall gene expression generated a metric that is comparable between samples. Correlations and associated p-values between groups of pre-CRT versus post-CRT and R versus NR was performed using Mann-Whitney U tests, p-values of <0.05 were considered significant. To account for multiple hypothesis testing, a Benjamini-Hochberg FDR of 0.1 was used to identify highly significant associations.

Immunohistochemistry

Details of the 6 antibodies (PD1, PD-L1, PD-L2, CTLA4, CD4, CD8), host species, clone, and dilatation are given in Supplemental Table 1. Immunohistochemistry (IHC) was performed automatically using a Benchmark XT/Discovery ULTRA Staining Module (Ventana Medical Systems, Inc., Tucson, AZ) using established protocols (42). In brief, protocols consisted of pretreatment with CC1 (pH 8.0), incubation with primary antibodies, and detection using a DAB-system (catalog No. 760-500, Ventana Medical Systems, Inc) including ultraview inhibitor, horseradish peroxidase, multimer chromogen, H2O2, and copper. In brief, sections were washed for 5 minutes (xylene x3, 100% ethanol x2, 95% ethanol x1, 70% ethanol x1, and PBS x1). Staining properties and specificity have been determined previously (Supplemental Table 1, (37,43–47)), which we additionally ascertained using negative and positive controls (Tonsil).

Microscopy and Quantification

For light microscopy, we captured images using an Olympus DP27 camera attached to an Olympus BX40 light microscope (Olympus America, Center Valley, PA). All markers were evaluated on tumor and non-tumor compartments and scored as positive vs. negative using established cut-offs (48–50). For CD4 and CD8 we additionally captured 4 images (high power field, 400x) and applied established image quantification tools. Briefly, segmentation of cells was achieved using threshold filters in combination with circularity and size cutoffs using “cell counter” and “analyze particle” plug-ins in Image J software (NIH, Bethesda, MD) (42). For statistical analysis of CD4 and CD8 staining of immune infiltrates, we took the average and median of four independent regions of interest. Differences in CD4 and CD8 T cell infiltrates between pre-/post-CRT samples were calculated using a t-test of changes with a null hypothesis of a difference of 0. p<0.05 was considered to be statistically significant. Correlations and associated p values between groups of pre-CRT versus post-CRT, R versus NR, and KP genotype versus no KP genotype were performed using Mann-Whitney U tests, p-values of <0.05 were considered significant.

Outcome analysis

We analyzed the association between R versus NR and KRAS/TP53 mutation genotype versus no KRAS/TP53 mutation genotype with progression-free survival using the Kaplan-Meier method. All statistical tests were performed using R version 3.5.2 and Prism 8 software (GraphPad, La Jolla, CA, USA).

Data availability

All BAMS for the matched pre and post-treatment tumors will be deposited in dbGAP (phs001829.v1.p1).

RESULTS

Chemoradiation does not increase TMB or neoantigen load

We assembled a cohort of 17 patients with locally advanced rectal carcinoma, of whom 9 were characterized pathologically as responders (R) and 8 as nonresponders (NR) following neoadjuvant CRT (Methods). Tumor genotype was unknown at the time of case identification. These patients had sufficient pre-CRT biopsy tissue and post-CRT surgical resection tissue available for multiple analytical pipelines including deep whole exome sequencing (Figure 1a). Demographic, treatment, and tumor characteristics are summarized in Supplementary Tables 2 and 3. All tumors demonstrated microsatellite stability. Median follow-up of the cohort was 47.1 months (range, 5.8-90.6). There were no local tumor failures. Overall, NR status was associated with reduced progression-free survival (PFS) compared to R with 5-year PFS 44% versus 100%, respectively (log-rank p=0.02) (Figure 1b). Median PFS for NR and R was 24.8 months and not reached, respectively.

Figure 1. Integrated molecular characterization.

(A) Sample inclusion and analytical workflow. (B) Progression-free survival by response. NR predicted poorer PFS compared to R with 5-year PFS of 44% versus 100% (log-rank p=0.02). (C) Mutational burden in cohort (paired t-test, p=0.4). Patients are ordered by response group (responders, nonresponders), with tumor mutation burden in decreasing order within each response category. (D) Neoantigen load in cohort (paired t-test, p=0.12). (E) Mutations in the cohort. Shown are the genes that were most commonly mutated as assessed by MutSig2CV analysis. CRT, chemoradiation; WES, whole exome sequencing; RNA, RNA-sequencing; IHC, immunohistochemistry; CR, complete pathological response

No statistically significant change in TMB before and after exposure to CRT was observed in our cohort (p=0.40, Figure 1c). A similar analysis of predicted neoantigen burden between pre- and post-CRT tumors also demonstrated no statistically significant change (p=0.12, Figure 1d). Neither pre- nor post-CRT neoantigen load were associated with treatment response (p=0.81, Supplemental Figure 1 and p=0.42, Supplemental Figure 2, respectively). We also found no difference in indel loads between pre- and post-treatment samples (p=0.20, Supplemental Figure 3). As has been previously demonstrated (51–54), the most frequently mutated genes pre- and post-CRT included KRAS, TP53, and APC (Figure 1e). Thus, global somatic mutations were not impacted by exposure to CRT in this cohort.

Presence of KRAS and TP53 co-mutation predicts resistance to chemoradiation

In evaluating differences in specific somatic mutations between R versus NR cases, we observed that NR tumors were enriched for concurrent KRAS and TP53 mutations (KP genotype) in contrast to R tumors (Fisher’s exact p=0.05, Figure 2a–b), as has been previously described (55–57). Notably, one pre-CRT KRAS-mutated tumor harbored a TP53 mutation post-CRT that was not detected in the pre-treatment tumor despite sufficient power to detect a mutation; this patient was also a NR (Figure 2c, Supplemental Figures 4–5), suggesting emergence of a radioresistant subclone. Given its association with NR, we next investigated the association between KP genotype and PFS. Patients with the KP genotype experienced reduced 5-year PFS (38%) compared to those without (90%, log-rank p=0.04, Figure 2d).

Figure 2. Co-mutation of KRAS and TP53 predicts resistance to chemoradiation.

(A) NR tumors were enriched for concurrent mutations in KRAS and TP53 genes compared to R (p=0.05, Fisher’s exact test). (B) Cancer cell fraction pre-/post-CRT for the KRAS and TP53 genes among the one R and six NR samples respectively. (C) Cancer cell fraction cluster plot for RC009 demonstrates the TP53 mutation in the post-treatment clones. (D) Patients harboring the co-KP genotype had poorer 5-year PFS (38%, log-rank p=0.04). CR, complete pathological response

Immune microenvironmental properties in rectal cancers treated with chemoradiation

To complement our investigation of tumor-intrinsic genomic properties discriminating response to CRT, we examined how transcriptional programs in the tumor or microenvironment were impacted by exposure to these therapies. Among the responders, there were 14 unique transcriptional programs significantly enriched in the pre-CRT samples and 1 unique transcriptional program significantly enriched in the post-CRT samples, with interferon alpha response genes enriched in both pre-/post-CRT samples (FWER p=0.00, Figure 3a). Among the NR, there were no unique significantly enriched transcriptional programs in the pre-CRT samples and there were 5 unique transcriptional programs significantly enriched in the post-CRT samples, with the angiogenesis and epithelial-mesenchymal transition (EMT) transcriptional programs enriched among both pre- and post-CRT samples (FWER p=0.00, Figure 3a).

Figure 3. Transcriptome and gene expression profiling identifies unique mechanisms behind radiation resistance among pathologic nonresponders.

(A) Gene set enrichment analysis demonstrated interferon alpha response genes enriched in both pre-/post-CRT samples in the R (FWER p=0.00), while angiogenesis and epithelial-mesenchymal transition genes were enriched in both pre-/post-CRT samples in the NR (FWER p=0.00). (B) The y-axis is an absolute quantification. Dotted lines represent individual paired patients (R vs. NR). Immune cell infiltrate significantly increased between pre-/post-CRT samples (t-test, p=0.04). (C) The y-axis is an absolute quantification, x-axis denotes immune cell subset populations with pre/post next to each other for each individual subset. Increased M2 macrophages were observed in post-CRT specimens amongst NR (Mann-Whitney U p=0.005, Benjamini-Hochberg FDR q=0.1), along with increased naïve B cells (Mann-Whitney U, p=0.03), monocytes (Mann-Whitney U, p=0.03), and resting mast cells (Mann-Whitney U, p=0.03). *** denotes significance per t-test; * denotes significance per Mann-Whitney U; ** denotes significance per Benjamini-Hochberg FDR

Given the immune-related transcriptional programs enriched pre-/post-CRT, we next examined immune cell infiltrates inferred from bulk transcriptome data (Methods). Total immune infiltrate levels were significantly higher in post-CRT specimens relative to their pre-CRT counterparts (p=0.04, Figure 3b). Overall, we observed significantly more naïve B cells (p=0.044), CD8 T cells (p=0.002), monocytes (p=0.01), M2 macrophages (p=0.002), and resting mast cells (p=0.0007) in the post-CRT tumor specimens. In contrast, there were significantly more memory B cells (p=0.04) and activated mast cells (p=0.006) in the pre-CRT tumor specimens (Supplemental Figure 6).

Interestingly, when limiting the analysis to NR pre-/post-CRT, we observed significantly more M2 macrophages (p=0.005, FDR q=0.1) in the post-CRT tumor specimens, as well as naïve B cells (p=0.03), monocytes (p=0.03), and resting mast cells (p=0.03), with significantly more activated mast cells in the pre-CRT specimens (p=0.04) (Figure 3c).

To complement bulk transcriptome analysis, we also evaluated immune infiltrate using immunohistochemistry for CD4 and CD8 T cells (Supplemental Table 4). The number of CD8 T cells trended toward a global increase between pre-CRT and post-CRT samples (p=0.47, Supplemental Figure 7). In the pre-CRT samples, there were more CD8 T cells in R compared to NR (p=0.14, Figure 4a) and complete responders (CR) samples had significantly more CD8 immune infiltration compared to NR (p=0.04, Figure 4a–b, Supplemental Figures 8–9).

Figure 4. Immune infiltrate analysis by immunohistochemistry demonstrates shifts in immune cell composition after chemoradiation.

(A) Among the pre-CRT samples, there were more CD8 cells among R compared to NR, and CR samples had significantly more CD8 cells compared to NR (Mann-Whitney U, p=0.04). (B) Representative images of a R with high CD8 infiltrate and NR with low CD8 infiltrate in pre-CRT specimens. (C) Among the pre-CRT samples, CR had increased CD4 immune infiltrate compared to NR (Mann-Whitney U, p=0.37). (D) Representative images of a R with high CD4 infiltrate and NR with low CD4 infiltrate in pre-CRT specimens. CR, complete pathological response; PR, partial pathological response; *significant p-value; ns, non-significant; Each individual dot represents the average of 4 independent regions of interest per patient.

Globally, CD4 infiltrate decreased between pre- and post-CRT, but this trend was not statistically significant (p=0.89, Supplemental Figure 10). Similar to CD8 T cells, NR trended toward having less CD4 immune infiltration compared to R (p=0.32, Figure 4c). When further breaking down response into CR versus partial responders (PR), CR appeared to have more CD4 immune infiltrate compared to NR (p=0.37, Figure 4c–d, Supplemental Figures 11–12). In summary, while IHC demonstrated significant differences in T cell infiltrate pre-CRT between R versus NR, clear shifts in immune infiltrate composition were observed after CRT in NR patients based on bulk transcriptome analysis.

DISCUSSION

To our knowledge, this is the first study to evaluate both genomic and microenvironmental changes at a primary rectal cancer tumor site exposed to preoperative CRT. Our data provide an opportunity to understand treatment-associated genomic changes between pre- and post-CRT specimens directly in patients. Tumor evolution has been previously studied primarily in the context of systemic cancer therapeutics in solid tumors (23,36,58–61), while most RT-based studies have examined candidate germline features or leveraged microarray data (11,12,62–75). Here we performed integrative comprehensive molecular characterization to dissect tumor and immune properties that track with CRT resistance.

Tumor mutation burden has been extensively studied and is suggested to be a marker of tumor-responsiveness to immune checkpoint blockade (18,19). It has been hypothesized that radiation may be able to increase TMB through its DNA-damaging mechanisms. Our data did not demonstrate an increase in overall mutational or neoantigen load after exposure to CRT. This finding is consistent with other pre- and post-matched tumor evolutionary assessments in the context of systemic therapy, particularly with cisplatin-based chemotherapy (23). Our data support the notion that chemotherapy or radiation are generally insufficient to prime the immune system by creating appropriate mutations or neoantigens (76,77).

While global genomic tumor properties were not clearly different between response groups, NR were more likely to harbor co-KRAS/TP53 mutations compared to R. The KP genotype has been previously suggested to be associated with radioresistance but the underlying mechanisms are poorly understood (55–57). Our observations in KP/NR cases suggest a previously unrecognized mechanism of immune suppression (Figure 3). We demonstrated that NR were more likely to express a M2 macrophage phenotype as well as enrichment for an EMT transcriptional program in the post-CRT specimens. The M2 phenotype is known to be anti-inflammatory, pro-angiogenic, and metastasis-promoting (78–80), while EMT plays a role in cancer metastasis and treatment resistance (81–85). Thus, KP/NR status may be associated with local immune escape during or after CRT. Of note, in our cohort without local recurrences, we found that NR/KP was associated with metastatic progression. Taken together, this suggests that KP/NR-associated local immune escape leads to distant metastatic disease and reduced PFS (Figure 1b, 2d). Thus, these tumors may benefit from novel neoadjuvant treatment approaches to reduce the risk of immune escape and metastatic seeding.

There are several limitations to this study. Small patient numbers make additional in-depth analyses and conclusions difficult, hence our findings need validation in larger, independent cohorts in diverse clinical settings. Many of our associations may be dependent on one another, as we do not have enough events to appropriately determine whether KRAS/TP53 genotype or pCR rate is more predictive of PFS through a multivariable regression. We rely on pCR as a biomarker of response, which has been called into question after preoperative CRT for rectal cancer (86) as pCR can vary and may be a function of time between end of CRT and surgical resection, although it has been used as a robust endpoint when evaluating novel systemic agents in other solid tumors (4). Some of our findings may be attributable to samples having higher or lower initial tumor burden; to overcome this issue, we performed purity/ploidy corrected molecular analysis through the ABSOLUTE algorithm (35) to account for differences in stromally admixed tumor specimens. We did not evaluate the impact of short-course preoperative radiotherapy nor other high-dose ablative radiotherapy schedules, which may elicit more mutagenesis and an immune response within the tumor microenvironment due to the higher dose per fraction during treatment (16,17,87,88). We also acknowledge that interpretation of in silico derived neoantigens from the mutations for each sample requires significant validation for improved interpretation. In addition, tumor spatial heterogeneity cannot be ruled out in this study as we do not have data from multiple areas of each tissue sample.

Overall, our study creates a path forward by leveraging molecular profiling for consideration of pre-operative CRT in patients with locally advanced tumors. This study also highlights the larger opportunity for additional investigations to elucidate novel mechanisms behind radioresistance across solid tumors.

SIGNIFICANCE/TRANSLATIONAL RELEVANCE.

Integrated tumor profiling of patient-matched rectal adenocarcinomas before and after neoadjuvant chemo/radiation therapy reveals insights into tumor evolution and treatment resistance mechanisms. The inability of neoadjuvant therapy to enhance tumor mutational burden coupled with poor response and local immune escape, particularly in KRAS/TP53-mutated tumors, warrant novel treatment approaches.