New insights into the responder/nonresponder divide in rectal cancer: Damage-induced Type I IFNs dictate treatment efficacy and can be targeted to enhance radiotherapy

Rectal cancer ranks as the second leading cause of cancer-related deaths. Neoadjuvant therapy for rectal cancer patients often results in individuals that respond well to therapy and those that respond poorly, requiring life-altering excision surgery. It is inadequately understood what dictates this responder/nonresponder divide. Our major aim is to identify what factors in the tumor microenvironment drive a fraction of rectal cancer patients to respond to radiotherapy. We also sought to distinguish potential biomarkers that would indicate a positive response to therapy and design combinatorial therapeutics to enhance radiotherapy efficacy. To address this, we developed an orthotopic murine model of rectal cancer treated with short course radiotherapy that recapitulates the bimodal response observed in the clinic. We utilized a robust combination of transcriptomics and protein analysis to identify differences between responding and nonresponding tumors. Our mouse model recapitulates human disease in which a fraction of tumors respond to radiotherapy (responders) while the majority are nonresponsive. We determined that responding tumors had increased damage-induced cell death, and a unique immune-activation signature associated with tumor-associated macrophages, cancer-associated fibroblasts, and CD8+ T cells. This signature was dependent on radiation-induced increases of Type I interferons (IFNs). We investigated a therapeutic approach targeting the cGAS/STING pathway and demonstrated improved response rate following radiotherapy. These results suggest that modulating the Type I IFN pathway has the potential to improve radiation therapy efficacy in RC.


Introduction
Colorectal cancer (CRC) is a devastating malignancy ranking as the third most common cancer diagnosis (10%) and second leading cause of cancer-related death in the United States (1). Historically, treatment for patients diagnosed with early-stage rectal cancer (RC) was total mesorectal excision surgery (TMES) (2). This surgery is particularly invasive as it requires partial removal of the rectum to eliminate the tumor, and some patients are rendered dependent on colostomy bags that signi cantly impact quality of life (3). As a result, physicians and scientists are exploring additional therapeutic combinations to treat patients with RC.
Neoadjuvant therapy reduces the risk for local recurrence following surgical excision (4). The current standard of care for RC is neoadjuvant short course radiotherapy (SCRT) or chemoradiotherapy (CRT) with the potential of some patients having to undergo excision surgery (5). Fortunately, a fraction of patients initially respond to preoperative therapy (e.g. tumor regression) making TMES unnecessary for this cohort. Consequently, an approach known as Watchful Waiting (WW) has emerged following preoperative RT or CRT to postpone surgery, thus sparing rectal function, and offers a promising alternative to surgical excision for preserving quality of life (6). Unfortunately, only 30% of patients demonstrate a clinical complete response (cCR) following preoperative therapy and are eligible for WW (7).
For TAMs, samples were processed by ltering out genes with low expression and performing a variance stabilizing transformation resulting in ltered genes assessed by Principal Component Analysis (PCA) with principal component (PC)-1 demonstrating a 65% variance and PC2 a 14% variance. The Euclidian distance was calculated based on the expression vectors between the untreated average to each irradiated sample and plotted against tumor size. The accension numbers for the RNA sequencing data Measuring phosphorylated gamma H2A.x Mice injected orthotopically with MC38-luc tumor cells received a single dose of 5 Gy targeted to the orthotopic tumor and were sacri ced three hours following RT. Tumors were processed for ow cytometry using uorescently conjugated anti-CD45 and anti-phosphorylated gamma H2A Histone Family Member X (H2A.x) antibodies.

Histology
Excised tumors xed in 10% neutral buffered formalin (Azer Scienti c, Morgantown PA, USA) were para n embedded, sectioned into 5-micron slices, and stained with anti-cleaved caspase 3 (Cell Signaling, Danvers MA, USA), anti-High Mobility Group Box I (HMGB1; Abcam, Cambridge UK), or anti-Calreticulin (Abcam), followed by secondary antibody (Vector Laboratories, Newark CA, USA) and counter stained with hematoxylin. Marker expression was determined by Aperio Image Scope algorithm. Spatial transcriptomics MC38-luc tumors harvested on day 11 (midway through SCRT) where 5-micron histological sections were adhered to spatially barcoded capture areas of a Visium Gene Expression Slide (10X Genomics, Pleasanton CA, USA). Following tissue placement, slide drying, and overnight incubation in a desiccator, the slide was depara nized, hematoxylin and eosin (H&E) stained, and imaged using a VS120 Slide Scanner (Olympus, Tokyo Japan) at 20X magni cation. Sequence ready libraries were constructed (10X Genomics, CG000407) with nal libraries sequenced on the NovaSeq 6000 sequencer (Illumina, Shanghai China) to obtain > 50,000 reads per spot and analyzed using the Seurat package in R (Uniform Manifold Approximation and Projection (UMAP) analysis, RNA assessment, bubble plots), and Loupe Browser (gene overlay).

IFN and IFNβ ELISA
MC38-luc tumor-bearing mice treated with either three doses of 5 Gy or the full SCRT dose were sacri ced on days eleven and fourteen, respectively. Tumors were harvested, weighed, snap frozen in Lysis II buffer, thawed, and manually homogenized. Homogenates were centrifuged at 300xG for ve minutes and supernatants analyzed using the Mouse IFN Beta High Sensitivity ELISA kit and Mouse IFN Alpha All Subtypes High Sensitivity ELISA kit (Pbl Assay Science, Piscataway NJ, USA). Concentrations normalized to grams of tumor tissue.
cGAMP treatment MC38-luc orthotopic tumor-bearing mice were injected with 10 µg/mouse 2'3'-Cyclic Guanosine Monophosphate-Adenosine Monophosphate (cGAMP; Millipore Sigma) intratumorally on day eight. Mice received SCRT, and immediately following each dose of RT were intravenously injected in the tail vein with 20µg cGAMP (in PBS). Tumor burden was monitored by IVIS or mice sacri ced on day twenty-six and primary tumors weighed.
Human CRC survival curve strati ed by STING1 Human patients with any stage of CRC from the Human Protein Atlas were strati ed based on high or low expression of stimulator of interferon genes (STING1) and survival was plotted. Survival plot terminated at 6 years.

Statistical analysis
Statistical analysis performed in GraphPad Prism 8 Software. BLI growth curves plotted as the geometric mean with standard deviation (SD) and survival determined by the Mantel-Cox test (p < 0.05). All other data presented as mean +/-SD. Signi cance for single comparisons determined by unpaired nonparametric Mann-Whitney T test and multiple group comparisons assessed by ordinary one-way Analysis of Variance (ANOVA) with multiple comparison post hoc tests.
RC patients treated with preoperative RT demonstrate a heterogeneous response; whereas a subset of patients' tumors initially respond to RT (20%), most patients are nonresponsive (10). To study this divide we utilized our established orthotopic model of RC and clinically relevant, targeted SCRT (8). Mice were injected with luciferase-expressing MC38 tumor cells intrarectally, titanium ducial clips were surgically inserted on opposing sides of the tumor on day eight, and tumors were targeted with SCRT from days nine through thirteen ( ve Gy per dose in ve consecutive fractions) (Fig. 1a). Mice were randomly grouped such that each group had tumors with equal geometric means (Fig. 1b). The responder/ nonresponder designation was determined by percent difference in tumor burden (BLI) as measured from the start of SCRT (day nine) to the end (day thirteen). Responders had a stable or negative percent change as opposed to nonresponders who demonstrated increased tumor burden (Fig. 1c) despite all animals having equal tumor growth prior to SCRT (Fig. 1d). A fraction of mice treated with SCRT (37%) respond to therapy and have reduced tumor burden, whereas the majority of mice, although treated with an identical dose of SCRT, demonstrate no reduction in tumor growth (Fig. 1e). Additionally, mice with radioresponsive tumors exhibited signi cantly enhanced overall survival compared to nonradioresponsive tumors (Fig. 1f).
Mice were retrospectively grouped as responders or nonresponders, and tumor sizes at individual timepoints during SCRT were analyzed to pinpoint when the difference in tumor burden emerged (Fig. 1g). Tumor burden was equal across all groups prior to the start of treatment on day six and remained consistent after the rst fraction of RT (day nine). However, by the third day of RT (day eleven), differences emerged, although it was not until the nal dose of SCRT (day thirteen) that the responding tumors were signi cantly smaller than the nonresponding. This demonstrated that the divide in response is not predetermined, but develops during the course of SCRT.
We demonstrated tissue targeting is precise in all animals and is likely not the cause of the varied response. γH2A.x is rapidly and transiently phosphorylated in response to double stranded DNA breaks (DSB). Phosphorylated (p)-H2A.x was signi cantly upregulated in all tumor tissue harvested three hours following a single dose of targeted SCRT compared to unirradiated controls and compared to adjacent tissue in irradiated individuals (Supplemental Fig. 1a-d). To further rule out that the magnitude of response is not dictated by RT targeting, MC38-luc cells were injected i.m. into the ank and the entire leg was irradiated with an identical 5 Gy by ve fractions SCRT schedule (Supplemental Fig. 2a). Although less clinically relevant, the advantage of this model is that the entire leg muscle was irradiated, making it unlikely the tumor would be outside the eld of treatment. In this scenario, the leg tumors also demonstrated a responder/nonresponder curve following SCRT (Supplemental Fig. 2b).
Intratumor hypoxia may alter effectiveness of RT as oxygen is essential for the generation of reactive oxygen species (ROS) that promote DSB (11). Although SCRT decreased the percentage of EF5 + (hypoxic) immune (CD45 + ) and nonimmune (CD45 − ) populations, there were no signi cant differences when strati ed based on responders/nonresponders (Supplemental Fig. 3). These data suggest that SCRT can reduce intratumoral hypoxia, but it is not preferentially reduced in responding tumors.
To investigate whether tumor cell clonality of the heterogenous MC38-luc cell line was contributing to the divided SCRT response, we generated three MC38-luc clones that were characterized in vitro (Supplemental Fig. 4) and subsequently implanted orthotopically where each clone also exhibited both responders and nonresponders to SCRT (Supplemental Fig. 5). To further generalize beyond MC38, we characterized clones of CT26-luc in vitro (Supplemental Fig. 6a-c) and selected one clone to test orthotopically in BALB/c mice. SCRT treated CT26-luc tumor-bearing mice similarly exhibited a responder/nonresponder divide (Supplemental Fig. 6d-f).
Collectively, our orthotopic responder/nonresponder model recapitulates the divide seen in patients following treatment. We next assessed what factors dictate this clinically-relevant response.
Although the quantity of immune in ltration does not differ, the phenotype of TAMs from responding tumors exhibit distinct polarization.
The e cacy of RT is largely mediated by the immune system (12). We hypothesized that the magnitude of response to SCRT is driven by either the quantity or quality of immune in ltrate. We focused on day fourteen as this was the earliest timepoint after SCRT where differences in tumor weight were identi ed (Supplemental Fig. 7a). There were no differences in the ratio of immune to nonimmune cells, nor in the number of speci c immune subtypes aside from monocytes (Supplemental Fig. 7b, c). Therefore, we performed a more in-depth analysis of the phenotype of in ltrating immune cells by bulk RNA sequencing of sorted immune populations. TAMs from the tumor homogenate of untreated, responding, and nonresponding tumors were selected and sequenced (Supplemental Fig. 8a) as they were the most abundant myeloid population in the TME (Supplemental Fig. 7c).
RNA sequencing results for the TAMs were plotted on a PCA plot. Untreated samples formed a tight cluster; however, the irradiated samples were dispersed throughout the plot (Fig. 2a). The Euclidian distance between each individual irradiated sample and the center of the untreated cluster were calculated. Values were correlated to tumor size resulting in a signi cant negative correlation (Fig. 2b, refer to Experimental Methods and Materials). The four irradiated TAM samples that were farthest away from the untreated average (and most genetically distinct) were also the four smallest tumors (e.g., responders) (Fig. 2c). In contrast, the samples that were most similar to the untreated in terms of genetic pro le were the largest tumors (e.g., nonresponders). This served as independent validation of our responder/nonresponder classi cation system. These results suggest that TAMs from nonresponding samples are genetically similar to untreated while TAMs from responding samples have a more distinct genetic pro le.
We performed pathway analysis on differentially expressed genes (DEGs) between responding and nonresponding TAMS and determined that responding TAMs were enriched for DAMP (HMGB1) and toll like receptor (TLR) signaling (Fig. 2d). Furthermore, we determined by gene set enrichment analyses (GSEA) that TAMs from responding tumors were enriched in ROS production (Fig. 2e). These data suggest TAMs from responding tumors exhibit a heightened response to damage.
Heightened levels of in ammation in TME of responding tumors in uences CAFs.
In addition to immune in ltration, the TME consists of other stromal cells. We sorted and sequenced CAFs from tumors ex vivo to determine whether the transcriptome of CAFs differed between responding or nonresponding tumors. Activated CAFs were classi ed as CD45 − CD31 − GFP − PDGFR + Podoplanin + Ly6C + (Supplemental Fig. 9a). Distinct genetic signatures were observed when DEGs between responding and nonresponding CAFs were assessed by hierarchical clustering (Fig. 2f). Our CAF dataset was compared to the 10,538 gene sets compiled by the BROAD Institute and analyzed by GSEA (Fig. 2g). Of the fty-eight signi cantly enriched pathways for the nonresponders, only nine of those pathways were related to immune regulation and function (16%) (Fig. 2g). However, of the thirty-seven signi cantly enriched pathways for the responders, twenty-eight were related to the immune system (76%), including the response to both Type I and Type II IFNs. These data demonstrate that, similar to TAMs, intratumoral CAFs from responding tumors exhibit a unique in ammatory signature.
MC38-luc cells puri ed from responding tumors display distinct phenotypes associated with damage and cell death.
We next investigated whether the transcriptome of RT responsive or nonresponsive tumor cells differed. MC38-GFP tumor cells were ow sorted from tumors ex vivo on day fourteen and RNA sequencing was performed (Supplemental Fig. 10a). Nonresponding tumor cells were similar to untreated tumor cells in terms of DEGs whereas the responding samples had a distinct genetic pro le (Fig. 3a). Furthermore, the magnitude of response was heightened in responding tumors, which exhibited almost 10-times more DEGs when compared to nonresponding tumors (1128 vs 131 DEGS when compared to untreated tumors, respectively: Fig. 3b).
We compared MC38-GFP sequencing results to a publicly available gene list for apoptosis and determined that cell death-associated genes were upregulated in responding tumor cells, but not in nonresponding tumor cells (Fig. 3c). Additionally, genes involved in DNA repair were speci cally enhanced in responding samples (Fig. 3c). We validated our RNA-sequencing results by performing IHC on day fourteen tumor sections for markers of cell death (cleaved caspase-3) and DAMP release (HMGB1 or calreticulin). Image analysis indicated that responding samples had signi cantly increased levels of cleaved caspase-3 and HMGB1, and a trending increase in calreticulin (Fig. 3d -f). Furthermore, increasing the amount of cell death by elevating RT doses (e.g., 8 Gy/fraction & 12 Gy/fraction) resulted in more responders (e.g., 60% & 80% respectively). (Supplemental Fig. 11).
Lastly, GSEA Pathway Analysis directly comparing responding tumor cells against nonresponding resulted in IFN response as the only signi cantly enriched hit (Fig. 3g). Type I IFNs are made in response to viral infection, or, more relevant in our model, in response to cell death and damage. Therefore, combined with the increases in cleaved caspase 3, HMGB1, and calreticulin determined by IHC, enrichment in Type I IFNs in the responding samples indicated elevated levels of damage and immunogenic cell death (ICD) compared to the nonresponding tumors.
Type I IFNs are increased in responding tumor regions compared to nonresponding and untreated tumors.
To con rm that Type I IFNs were elevated primarily in irradiated tumor tissue, and not in adjacent nonmalignant regions, we performed spatial transcriptomics on irradiated tumor sections and classi ed the different tissue regions based on UMAP clustering (Fig. 4a, Supplemental Fig. 12a,b). We identi ed regions of normal tissue (regions 1-4, 7-9) based on epithelial cellular adhesion molecule (EpCAM) expression and malignant regions based on histological assessment along with a lack of EpCAM expression and positive cytokeratin (Cdk6) staining (region 0; Fig. 4b, Supplemental Fig. 12b). Expression of Type I IFN mRNAs were predominantly localized to the tumor region (Fig. 4c). To complement this nding, we assessed various genes associated with responsiveness to Type I IFNs ( Table 1) and illustrated that the in ammatory response to Type I IFNs were similarly colocalized only within the tumor region (Fig. 4d). We referenced the transcriptomes of the cells in this region and determined that aside from tumor cells, the major immune cells in this section were dendritic cells (DCs; Itgax/ Cd11c) (Supplemental Fig. 12b) and CD8 + T cells (Fig. 4e), both of which are immune populations in uenced by Type I IFNs (13).
We quanti ed the intratumoral concentration of Type I IFN protein in responding and nonresponding tumors by ELISA. Intratumoral IFNβ protein concentrations were signi cantly increased in responding tumors compared to nonresponding and untreated tumors on day eleven (Fig. 4f). These values decreased by 50% on day fourteen, however, irradiated tumors still demonstrated a signi cant increase compared to untreated (Fig. 4g). Likewise, intratumoral IFN concentrations in the responders demonstrated signi cant increases when compared to untreated tumors on day 11 (Fig. 4h) and were still trending higher at day 14 (Fig. 4i). The amount of Type I IFNs negatively correlated with tumor size on days eleven and fourteen, where smaller tumors (i.e., responders) had increased IFN /β concentration compared to larger tumors (i.e., nonresponders) (Fig. 4j, k). These results con rm the importance of Type I IFNs, especially early on, in promoting a response to SCRT.
CD8 + T cells are essential for driving a response and are heavily in uenced by Type I IFNs. Figure 4 identi ed abundant CD8 + T cells in the tumor region that colocalized with Type I IFNs. Depletion experiments determined that CD8 + T cells were essential for generating a response to SCRT (Fig. 5a, b).
Based on these data, we hypothesized that responder tumors maintained higher quality CD8 + T cells compared to nonresponders. To test this, we sorted CD8 + T cells from irradiated or untreated tumors and performed bulk RNA sequencing (Supplemental Fig. 13). Unbiased hierarchical clustering of DEGs grouped CD8 + T cells from nonresponding tumors with those in untreated tumors, suggesting these two groups were similar to each other. In contrast, CD8 + T cells from responder tumors displayed clear differences in gene expression from both nonresponders and untreated CD8 + T cells (Fig. 5c). Pathway analysis revealed only minor differences between the nonresponding and untreated CD8 + T cells, however striking differences were observed between responder and untreated CD8 + T cells (Fig. 5d). These results included pathways related to antigen experience, cytokine and chemokine signaling, response to DAMPs, cell cycle regulation, and fate. Collectively, these results indicated a more activated T cell transcriptome in responding tumors.
To explore the impact of Type I IFN signaling on T cell effector status, we treated tumor-bearing mice with an anti-IFNAR neutralizing antibody followed by SCRT and harvested CD8 + T cells for bulk RNA sequencing. The transcriptome of CD8 + T cells treated with IgG exhibited signi cant differences from the T cells harvested from the IFNAR-treated mice (Fig. 5e). The top 30 pathways that were signi cantly downregulated when Type I IFN signaling was blocked by the IFNAR antibody are hallmarks of the Type I response, including regulation and signaling of the Type I IFN pathway, viral response genes, and cytokine production (Fig. 5f). Furthermore, CD8 + T cells from IFNAR treated mice grouped more closely with nonresponders while responding CD8 + T cells formed their own distinct cluster (Fig. 5g). These data suggest that heightened Type I IFN signaling from responding tumors promotes an activated/effector-like T cell phenotype.
Type I IFN drives the responder tumor phenotype following SCRT. Figure 5 demonstrated that blockade of Type I IFN signaling renders CD8 + T cells similar to nonresponders/untreated cells. We postulated this blockade will result in a poor response to SCRT and predominately nonresponder tumors. Neutralization of Type I IFN signaling with an IFNAR blocking antibody (Fig. 6a) resulted in 100% of IFNAR-treated mice being nonresponsive to SCRT (Fig. 6b) with larger tumor burden upon sacri ce on day twenty compared to the IgG-treated controls (Fig. 6c, d). Our results highlight the importance of Type I IFN in dictating SCRT e cacy.
Activating the cGAS-STING pathway increased the number of responders.
There are various mechanisms by which Type I IFN can be triggered in response to damage. For example, extracellular DNA is recognized by over ten different cytosolic receptors, one of which is cGAS, which upon binding cytoplasmic DNA becomes catalytically active and generates cGAMP. cGAMP binds STING and causes translocation to the golgi resulting in Type I IFN secretion (14). Using a human CRC dataset collated by the Human Protein Atlas we determined that patients with high STING expression had improved overall survival compared to patients with low STING expression (Supplemental Fig. 14) justifying a therapeutic intervention to target this pathway.
Tumor-bearing mice were treated with cGAMP intratumorally on day eight followed by intravenous administration daily throughout SCRT (Fig. 7a), and mice sacri ced on day twenty-seven. SCRT-treated tumors demonstrated the typical 30-40% response rate, however mice treated with SCRT in combination with cGAMP demonstrated an 80% response rate where eight of the ten individuals had minor, if any, residual primary tumor (Fig. 7b, c, d). These results demonstrate that cGAMP and SCRT combinatorial therapy signi cantly improve localized response compared to SCRT alone.

Discussion
The data presented demonstrate that SCRT induces a responder/nonresponder phenotype in multiple murine models of RC, indicative of the response observed clinically. Our ndings suggest that ICD drives acute Type I IFN signaling resulting in a more activated subset of CD8 + T cells and TAMs in the responding samples compared to the nonresponding samples. Although not clinically relevant, increasing the dose to 12 Gy/fraction results in an 80% response rate, compared to the 40% response rate when 5 Gy is administered. This experiment informed us that it was possible to adjust the ratio of responders, and that radiation-induced cell death may be a contributing factor. Alternatively, and more clinically relevant, we determined that amplifying the cGAS/STING signaling pathway by therapeutically agonizing cGAS results in an increase in the percentage of responders to SCRT.
We demonstrated that SCRT e cacy was signi cantly blunted when animals were treated with a blocking antibody against Type I IFN signaling. Type I IFNs encompass a large family of structurally related monomeric pleiotropic cytokines 18 where IFN and IFNβ are the most well-characterized (15). In addition to cellular damage (16), viral or bacterial invasion, extracellular self-nucleic acid (tumor-secreted DNA)induced cGAS / STING signaling, or binding of other non-nucleic acids (DAMPs) via TLRs (14) all drive Type I IFN production. Type I IFNs activate several immune subsets, including T cells, via the IFNAR to stimulate immunity following RT-induced cellular damage. For example, Type I IFNs increase cytotoxicity and cytokine production of CD8 + T cells (17) and promote durable T cell memory responses following viral infection (18). These studies extend beyond responses to viruses as Type I IFNs have recently been accredited with priming an immune response against tumor cells, especially in the context of RT (19). Many of these effects are mediated by triggering of cGAMP/STING. This is further emphasized by data demonstrating that heightened STING expression in human CRC correlated with increased overall survival (Supplemental Fig. 14).
A key question is why is there increased ICD in a subset of tumors even through all received similar RT doses? We ruled out differences in oxygenation status, and the clonality experiment con rmed that genetic predispositions of individual tumor cells were not a driver of response rate. Further genetic analysis of the tumors by RNA sequencing determined that the MC38 tumor cells harvested ex vivo from irradiated tumors are undergoing DNA damage and apoptosis, however this may be a readout of RT e cacy, rather than an indicator of RT-sensitivity. Further experimentation to determine what causes increased tumor cell death in a subset of tumors is needed to entirely elucidate this phenotype, however it is possible that epigenetic differences emerge when tumor cells are implanted and irradiated in vivo.
In accordance with the robust literature surrounding Type I IFN signaling, Type I IFNs were approved by the Food and Drug Administration (FDA) for cancer treatment in 1986. However, to our knowledge, no current clinical trials explore the combination of IFN treatment alongside SCRT for RC. Based on the data presented here, we propose that quanti cation of intratumoral Type I IFN protein concentrations and DAMP secretion of irradiated RC biopsies could serve as a predictive biomarker to determine who is likely to respond preoperatively. Additionally, our data supports a non-randomized Phase II clinical trial for patients with locally advanced, Stage II/III RC. The goal of this trial would be to demonstrate that cGAMP in combination with the standard of care treatment results in decreased tumor size and invasion at the time of resection, or alternatively that combination treatment results in a higher rate of responders who can delay TMES, compared to patients treated with CRT alone.