Clinical and genetic factors associated with tumor response to neoadjuvant (chemo)radiotherapy, survival and recurrence risk in rectal cancer

Rectal cancer poses challenges in preoperative treatment response, with up to 30% achieving a complete response (CR). Personalized treatment relies on accurate identification of responders at diagnosis. This study aimed to unravel CR determinants, overall survival (OS), and time to recurrence (TTR) using clinical and targeted sequencing data. Analyzing 402 patients undergoing preoperative treatment, tumor stage, size, and treatment emerged as robust response predictors. CR rates were higher in smaller, early‐stage, and intensively treated tumors. Targeted sequencing analyzed 216 cases, while 120 patients provided hotspot mutation data. KRAS mutation dramatically reduced CR odds by over 50% (odds ratio [OR] = 0.3 in the targeted sequencing and OR = 0.4 hotspot cohorts, respectively). In contrast, SMAD4 and SYNE1 mutations were associated with higher CR rates (OR = 6.0 and 6.8, respectively). Favorable OS was linked to younger age, CR, and low baseline carcinoembryonic antigen levels. Notably, CR and an APC mutation increased TTR, while a BRAF mutation negatively affected TTR. Beyond tumor burden, SMAD4 and SYNE1 mutations significantly influenced CR. KRAS mutations independently correlated with radiotherapy resistance, and BRAF mutations heightened recurrence risk. Intriguingly, non‐responding tumors with initially small sizes carried a higher risk of recurrence. The findings, even if limited in addition to the imperfect clinical factors, offer insights into rectal cancer treatment response, guiding personalized therapeutic strategies. By uncovering factors impacting CR, OS, and TTR, this study underscores the importance of tailored approaches for rectal cancer patients. These findings, based on extensive analysis and mutation data, pave the way for personalized interventions, optimizing outcomes in the challenges of rectal cancer preoperative treatment.


| INTRODUCTION
With a global incidence of approximately 1.9 million cases annually, colorectal cancer (CRC) is the third most common cancer type. 1 One third of all CRCs are located in the rectum.Improved surgical techniques and the use of preoperative radiotherapy (RT) or chemoradiotherapy (CRT) have led to considerably fewer local recurrences and better survival in these patients. 2Preoperative RT/CRT is usually added to allow radical surgery of advanced tumors and/or to lower local recurrence rates in less advanced cases. 3The response to RT/CRT is highly heterogenous and in up to 30% of patients, a complete pathological response (pCR) is seen at surgery. 4,57][8][9] Because of this excellent response in some patients and the negative consequences surgery may have, 10,11 a third reason to provide RT/CRT is to avoid surgery if a clinical complete response (cCR) is seen and instead enroll the patient in a watch-and-wait protocol (W&W). 8,12e possibility to accurately identify the responders at time of diagnosis could enable a personalized treatment approach for rectal cancer patients.Small tumor size and early tumor stage have been associated with CR, 13,14 but to understand the heterogenous tumor response, studies of the tumors on genetic and transcriptomic levels are warranted.Much effort has been put into this and KRAS mutations have frequently been linked to RT-resistance, [15][16][17][18][19] but great heterogeneity between studies have been noted.Mutations in BRAF and SMAD4 have in one study been associated with poor response to CRT and decreased disease-free survival (DFS). 20Genes connected with pCR (or cCR) are however less studied, and, although some molecular biomarkers have been suggested as predictors of response to CRT, none have yet reached the clinic. 19,21,22For a recent review, see De Mattia et al. 23 Studies aiming to identify predictors of treatment response are often challenged by the limited number of CRs and the need for untreated tumor tissue, that is, the small diagnostic biopsies, for analyses.
In this study, we have combined clinical and targeted sequencing data in a large population-based rectal cancer cohort to assess what determines complete tumor response to pretreatment.We also investigated which factors impact overall survival (OS) and time to recurrence (TTR) after primary treatment.Knowledge of the latter could help when deciding whether additional treatments are needed, to prevent recurrence.

| MATERIALS AND METHODS
All patients diagnosed with non-metastatic rectal cancer during the period 2010-2018 and living in two regions in Sweden (Uppsala and Dalarna, total population 630,000 in 2015) were identified and considered for study inclusion.Patients receiving preoperative shortcourse RT (scRT), CRT or scRT followed by chemotherapy (scRT+CT)   with either a delay to surgery, enrollment in a W&W program, or with evaluable response using magnetic resonance imaging (MRI) were selected for treatment response, OS and TTR analyses.

| Clinical and pathological staging and treatment
All patients had an invasive rectal adenocarcinoma, located in the most distal 15 cm of the bowel.As previously described, 24 staging included clinical investigation, routine blood samples, computed tomography of the liver, abdomen and lungs and MRI of the pelvis.All patients were discussed in multidisciplinary team meetings prior to treatment decision.Both staging and treatment followed the national care programs from 2008 and 2016: early tumors with low risk of recurrence were assigned to upfront surgery, tumors with intermediate recurrence risk to scRT with immediate or delayed surgery and locally advanced, high-risk tumors to CRT (25 Â 2 Gy or 28 Â 1.8 Gy with simultaneous fluoropyrimidine treatment) with delayed surgery.In elderly unfit patients, scRT was used as an alternative to CRT.During the study period, three ongoing clinical trials influenced the therapy given: Stockholm III, 25 RAPIDO 4 and LARCT-US. 26In the Stockholm III trial, patients were randomized to either scRT (5 Â 5 Gy) with immediate or delayed surgery or long-course RT (25 Â 2 Gy).In the RAPIDO trial, patients were randomized to the standard of care CRT with delayed surgery (with adjuvant chemotherapy) or the experimental treatment scRT followed by chemotherapy (six cycles of CAPOX).In the LARCT-US trial, patients were treated with scRT followed by four cycles of CAPOX and delayed surgery.
Clinical data on all patients were extracted from the Swedish Colorectal Cancer Registry (SCRCR) and completed by a retrospective review of the medical records.Upper normal was used as a cut-off in laboratory parameter blood analyses: 3.8 μg/L for carcinoembryonic antigen (CEA) and 5 mg/L for C-reactive protein (CRP).A hemoglobin (Hb) value of ≤110 g/L was defined as anemia.Tumor size was the maximum tumor length in centimeters determined by MRI.In cases with missing information on clinical staging, a radiologist (N.K.H.) re-examined the original MRI images.Staging followed the UICC TNM-7 classification.
Tumor regression was classified as complete (pCR/cCR) if the tumor had disappeared completely after surgery (ypT0N0) or was in complete clinical remission on posttreatment palpation, endoscopy, and MRI 3 months after the last treatment and remaining for at least 12 months.If the tumor was still present after treatment, tumor regression was classified as non-complete (non-CR).Carcinoma in situ after surgery was considered as pCR in the analyses (applied to one patient).For library preparation, 10 ng of UDG treated DNA samples was used and sequencing was performed on S5 550 chips using the Ion S5 Sequencing Systems™ (Thermo-Fisher Scientific, USA).The Oncomine™ Tumor Mutation Load assay (Thermo-Fisher Scientific, USA) covering 1.7 Mb across 409 cancer-driver genes (Supplementary Table 1) and allowing for tumor mutational burden assessment was used for targeted sequencing.Variant calling analyses were performed in the Ion Reporter™ software (Thermo-Fisher Scientific, USA) with the hg19 human genome as reference.Somatic single nucleotide variants (SNV) and insertions and deletions (INDEL)   were called according to Oncomine Tumor Mutation Load Regions v2.1 with a minimum variant Phred-score average quality of 10 and 20 and a minimum coverage threshold of 60 and 15, respectively.
For both variant types a minimum variant allele frequency (VAF) threshold of 0.125 was used and a strand bias under 80% was allowed.Somatic hotspots were detected with the Oncomine Tumour Mutation Load Hotspots v2.2 software with a minimum average base quality of 10, a minimum coverage threshold of 15, a minimum VAF of 0.1 and no more than 80% support on each strand.
Oncomine Tumor Mutation Load Assay Annotations v1.2 plus the Oncomine Extended filter chain were used to annotate somatic variants and filter out germline variants present in 500Exomes, ExAC or USCS Common SNPs databases.Tumor mutation burden (TMB) was calculated post variant calling using the TMB Non-germline Mutations v4.0 filter that included exonic non-synonymous SNV and INDEL variants with a minimum base coverage of 60 and VAF of 0.15.Samples with deamination score >30 were excluded from TMB reporting.Biopsies with a mean read depth of ≥800 were included in the analyses.The sequencing coverage and quality statistics for each sample are summarized in Supplementary Table 2.

| Hotspot detection in KRAS, NRAS, BRAF and PIK3CA
For biopsies with TCC <10% or of too small size (no representative tumor left for diagnostic purposes), hotspot data were retrieved from the patients' medical records, if sequencing had been done as part of the clinical routine (n = 71).In clinical routine, pyrosequencing (prior to November 2014) 27 and later next generation sequencing (NGS) using NGS HaloPlex gene panels 28 were available.Mutational status for hotspot positions in KRAS and NRAS codons 12, 13, 58, 59, 61, 117 and 146 and BRAF codon 600 and (since 2016) PIK3CA codons 542, 545, 546, 1021, 1043, 1044 and 1047 have been reported.For targeted sequencing cases with a lower mean read depth (<800), hotspot information for BRAF, KRAS, NRAS and PIK3CA was retrieved if sufficient targeted sequencing data on these genes was available according to the parameters above (n = 11).The sequencing coverage and quality statistics for these samples are summarized in Supplementary Table 3.For cases where fresh frozen material was available (n = 38), frozen tissue blocks were prioritized over FFPE material for whole genome sequencing (WGS) on a DIPSEQ platform (BGI, Shenzhen).Reads were mapped with Sentieon Genomics software and somatic mutations were called using multiple accelerated tools. 29Only data for the above-mentioned hotspots were included in the following analyses for these frozen samples.The sequencing coverage and quality statistics for these samples are summarized in Supplementary Table 4.

| Statistics
Patients were divided into groups based on the type of sequencing data available (Figure 1B): (i) patients with mutation data from the targeted sequencing panel (n = 216) and (ii) patients with hotspot data (either from the Oncomine panel, WGS or clinical routine NGS or pyrosequencing, n = 336).These groups were analyzed separately.If the biopsy was too small for sectioning or had TCC <10% and no hotspot data was available from clinical routine or the WGS, the patient was excluded from any mutation analyses (n = 66).Clinical data (including laboratory parameters and tumor specifics) and mutational data and their associations to CR/non-CR were investigated in all three study groups.Continuous variables such as age, tumor size and tumor level were transformed into categorical variables.Clinical tumor (cT) stage was divided into two groups: cT1-3a and cT3b-4b.In accordance with our previous study, 13 tumors in stage cT3a were grouped with stage cT1-2 tumors due to the difficulty in radiologically distinguishing cT2 from cT3a 30  Binary logistic regression with backward elimination was used to estimate odds ratios (OR) and 95% confidence intervals (CI).p < .05 was used as an inclusion criterion for covariate selection between steps.
Variables with >10% missing data were excluded from the multivariable analyses.Factors directly correlated to another factor, for example, treatment and OTT were excluded to avoid collinearity in the analyses.
CIs not crossing one were considered statistically significant.
Receiver operating characteristic (ROC) analysis was used to evaluate the ability of the logistic regression models to accurately predict CR.For this purpose, the fitted values of the four different models and the outcome measures CR and non-CR were used to obtain ROC curves and area under the curve (AUC) values with bootstrap repetitions set to 100,000 and 95% CI.OS was defined as the time from rectal cancer diagnosis to death and TTR as the time from surgery (postoperative patients) or 6 weeks after last treatment (W&W patients) to distant or local recurrence.For patients in the W&W group, any instance of locoregional or distant recurrence was an event for the TTR calculation; however, regrowth followed by successful radical (R0) surgery was not considered as an event.Patients alive or recurrence-free on February 05, 2022, were censored.Non-operated patients with tumor response evaluable by MRI, but with remaining tumor after treatment (non-CR) were excluded from the TTR analyses.OS and TTR were assessed using the Kaplan-Meier method for visualization with log rank test for group comparisons.Factors significantly associated with OS/TTR were included in Cox regression models to estimate hazard ratios (HR), CIs and p-values for each predictor.In univariable analyses, combined cTN stage, cT stage, tumor size, CRP, CEA, and treatment (or OTT) were associated with tumor response, with small, early-stage tumors being more likely to reach CR (Table 1).

| Patient material
In a multivariable analysis and, in line with our previous report, 13 treatment and tumor size were independently associated with tumor response with decreasing odds of reaching CR with increasing size (OR = 0.26, 95% CI [0.09-0.78]for tumors of 3-5 cm and OR = 0.15, 95% CI [0.05-0.49]for tumors of >5 cm, with tumors <3 cm in size as reference, Table 2).Patients treated with CRT or scRT+CT had odds of reaching CR that were 4.99 (95% CI [1.86-13.39])and 7.47 (95% CI [3.00-18.64])times those of patients treated with scRT (Table 2).
A CR was associated with improved OS (Supplementary Figure 1A) and TTR (Supplementary Figure 1E), with >90% of CR patients being alive and recurrence-free after 5 years, compared with $70% with non-CR to treatment (p < .001).Treatment was also associated with OS, with scRT treated patients having significantly worse OS compared to CRT and scRT+CT treated patients ( p < .001,Supplementary Figure 1B).This can mainly be explained by more elderly and frail patients being allocated to scRT.The median age among patients receiving scRT was 75 years (Supplementary Table 5) and patients that were ≥80 years at diagnosis had significantly worse survival than younger patients ( p < .001,Supplementary Figure 1D).
See brick plot in Figure 2, all mutations are presented in Supplementary Data.
In this exploratory cohort, all mutations in the 12 genes were considered.In univariable analyses, none of these mutated genes were associated with tumor response (Table 3).In contrast, and similar to the entire cohort, treatment, and factors reflecting tumor burden, such as cTN stage, cT stage, size and pretreatment CEA level, were associated with treatment response (Table 1).Small and early-stage tumors (stage I or cT1-3a) were more likely to reach CR after treatment (Table 1).
In (n = 25, 12%) signaling pathways were not associated with tumor response (Table 3).When these pathways were included in a multivariable analysis, only treatment, tumor stage and size were predictive of CR (Table 2).In the subgroup of patients receiving scRT+CT (n = 64), the WNT pathway was predictive of outcome with altered cases having increased odds of reaching CR (OR = 6.27, 95% CI [1.26-31.17]).
In the WNT pathway, 95% of the cases had a mutation in the APC gene.
In the targeted sequencing cohort, CR was associated with better OS (p < .001,Supplementary Figure 3A).Older age and scRT treatment were both associated with worse OS (p < .001for both, Supplementary Figure 3B,C).In a multivariable Cox regression analysis including these factors, CR and age remained associated with OS (Supplementary Figure 2B).CR was associated with better TTR, with >90% of patients being long-term recurrence-free ( p = .003,Supplementary Figure 3D).In addition, an APC mutation was independently associated with better TTR (Figure 3A and Supplementary Figure 3F), with 80% of the patients being recurrence-free after 5 years.
Mutations in SMAD4 were associated with worse TTR ( p = .042,Supplementary Figure 3E); however, this finding did not remain statistically significant when adjusting for tumor remission and APC status in a Cox regression analysis (Figure 3A).

| Hotspot cohort (n = 336)
The hotspot cohort included targeted sequenced patients (n = 227), patients with small and low TCC biopsies sequenced in clinical routine positions (six in BRAF, two in KRAS and eight in PIK3CA) and were considered wild type in these analyses.In the final cohort, the incidence of hotspot mutations in the different genes were: 45% KRAS, 9% PIK3CA, 5% NRAS and 3% BRAF V600E (Table 3).
Tumor size, baseline CEA, treatment and KRAS status were determinants of CR in both univariable (Tables 1 and 3) and multivariable analyses (Table 2).A hotspot mutation in KRAS more than halved the odds of reaching CR (OR = 0.43, 95% CI [0.21-0.89]).Small tumor size and the addition of chemotherapy increased the likelihood of CR.
Of the patients with KRAS mutations, 25 had a mutation in codon 13 (of which 22 were G13D) and one of these patients (4%) responded completely.In comparison, CR was achieved among 12% of the patients with other KRAS mutations (15/127).However, this finding was not statistically significant ( p = .473).The most common type of KRAS mutation among CR patients was a codon 12 alteration (44%, 7/16: four G12D and three G12V).
In accordance with the complete study cohort, tumor remission, treatment, baseline CEA and age were associated with OS (Supplementary Figure 4A-D and Supplementary Figure 2C).CR after treatment was associated with better TTR ( p < .001,Supplementary Figure 4E) and BRAF V600E with significantly worse TTR ( p < .001,Supplementary Figure 4F).In a multiple Cox regression analysis, BRAF V600E remained associated with worse TTR and CR with better TTR (Figure 3B).

(A) (B)
F I G U R E 3 Cox regression analyses with hazard ratios and 95% confidence intervals for time to recurrence in (A) targeted sequencing cohort and (B) hotspot cohort.Non-operated patients judged non-CR by MRI were excluded from the TTR analyses (targeted sequencing cohort: n = 6, hotspot cohort: n = 7).Log-transformed scale.CR: complete remission.

| TTR analyses in non-CR patients (n = 179)
For all three study groups the recurrence rate was <10% if the tumor responded completely to therapy.Factors associated with TTR were investigated in the subgroup of patients with targeted sequencing data and tumor left after treatment (non-CR patients, n = 179) and showed that patients with a mutation in the WNT pathway (including APC) were less likely to relapse (Supplementary Figure 5A,B).A mutation in BRAF was associated with worse TTR (Supplementary Figure 5C,E) with a 50% relapse rate after 3 years.Initially small non-CR tumors (≤5 cm in size) had shorter TTR compared with initially large tumors (>5 cm, Supplementary Figure 5D).After 3 years, 40% of non-respondent patients with a small primary tumor had a recurrence compared with 20% among patients with large tumors.Tumor size and a WNT pathway mutation remained significantly associated with TTR in multivariable Cox regression analyses (Supplementary Figure 5F,G).
Further analyses were performed to investigate why small non-CR tumors were more likely to relapse.Mutated TP53 was significantly more common among initially small non-CR tumors (≤5 cm) compared with large non-CR tumors (>5 cm, p = .026,Supplementary Table 6).TP53 mutations were more common among small tumors ( p = .021and p = .026in the full targeted sequencing cohort and among the non-CR patients, respectively), but as stated above, not linked to tumor response.Among non-CR patients, tumor level influenced the CEA value at diagnosis and extramural vascular invasion (EMVI) status, with smaller proportions having high CEA and EMVI+ among low tumors (Supplementary Table 6).Moreover, several genetic differences were seen according to tumor level: NRAS, SMAD4, FBXW7 and TCF7L2 mutations were more common among high tumors compared with tumors located in the low/mid rectum.
Mutations in the TGFβ pathway were more common in high tumors (27% in high rectum compared with 10% in low/mid-rectum, p = .015).

| Accuracy of prediction models
The ability of the four different logistic regression models to predict CR was evaluated through ROC analysis.In each group, the factors with independent association with tumor remission (Table 2) were included in the separate models and the respective AUC values, the average height of the ROC curves, were interpreted as a measure of the accuracy of the predictive capacities

| DISCUSSION
In this cohort of locally advanced rectal cancers from a defined Swedish population, treatment had a strong correlation with outcome, with total neoadjuvant therapy (TNT, scRT+CT) having 7.5 times higher odds of reaching a CR compared with RT only.Large tumors (>5 cm) were 10 times less likely to disappear compared with small tumors (<3 cm).When data on the hotspot mutations in KRAS, NRAS, BRAF and PIK3CA genes, presently of greatest clinical value in CRC, were included, treatment and tumor size remained associated with response, with the addition of high levels of CEA and a KRAS mutation being negative predictors of CR.Treatment and stage (rather than size) had strong impact on outcome also in the smaller targeted sequencing cohort with additional independent information from mutations in KRAS, SMAD4 and SYNE1, out of the 12 most frequently mutated genes.Analyzed pathways were not independently associated.7][8][9] In addition to response, a mutation in APC was associated with a longer TTR compared to wild type.In the hotspot cohort, a BRAF V600E mutation meant an independently increased risk of recurrence.Several mutations were, however, associated with relapse risk in patients not responding completely, that is, in those with a higher need for adjuvant treatment to reduce the risk of relapse.
8][19] In a large study on 1886 stage II-III rectal cancers treated with CRT, KRAS mutations were, however, not associated with pCR, but with worse OS. 31 Chatila et al. 32 reported an association between mutations in KRAS and shorter DFS but did not find a correlation with CR.Here, KRAS mutation status did not impact OS or TTR, but was associated with a decreased chance of CR.A decreased chance of CR in KRAS mutated tumors was also found in a systematic review; in fact, KRAS mutation was the only mutation associated with CR. 23 It was, however, the most studied gene, and great heterogeneity was found between the different studies.It has previously been reported that specific KRAS codons may have different impact on RT-resistance, 17 with codon 13 mutations being especially resistant to CRT treatment.In line with this, we report fewer CR among patients with a KRAS codon 13 mutation (4%) than among patients with other KRAS mutations (12% CR), this finding was, however, not statistically significant.SYNE1 has not previously been reported as associated with tumor remission.It is one of the largest genes in the human genome that encodes an 8797 amino acid-long protein and is frequently mutated in CRC. 33,34The mutations in SYNE1 seen here were mostly SNVs which do not suggest gene inactivation and they were not recurrent, indicating that no hotspot is activated.For these reasons, the relevance of these mutations is unclear and further investigation is warranted.
Associations between mutations in BRAF and SMAD4 and resistance to CRT and shorter DFS (called progression-free survival in the publication) have been reported, albeit not universally. 20,23In line with this, a mutation in BRAF meant worse TTR in this study; however, based upon few cases (n = 10) as these mutations are rare in rectal cancer.In the present study, a SMAD4 mutation was associated with increased odds of CR (OR = 6) compared with wild type.Notably, only 15 patients (7% of the targeted sequencing cohort) had an altered SMAD4, making it difficult to draw firm conclusions.Most mutations in the SMAD4 gene were SNVs and four (all non-CR) were in a known hotspot. 35crosatellite instability (MSI) has, in one study, been associated with reduced pCR rate after CRT in rectal cancer 36 but no association could be found when literature was systematically reviewed. 23In the present study, TMB was used in the targeted sequencing cohort as an indirect measure of MSI but excluded from response analyses since only two cases had high TMB (both non-responders to scRT+CT).We hypothesize that estimation based upon TMB underestimates MSI slightly, as we expected a frequency around 2%. 37 Of the known driver genes in CRC, APC was the most frequently mutated.Wild type APC has been associated with worse survival than when an APC mutation is present, 38 like our finding of a higher risk of recurrence if APC was not mutated.In the subgroup of patients receiving scRT+CT, a WNT pathway mutation was associated with increased odds of reaching CR.This was, however, not seen in the targeted sequencing cohort when all treatments were explored and may be coincidental.We report a lower APC mutation frequency than expected (56% vs., e.g., 78%-80% in the MSKCC-rectal and TCGA-READ cohorts).This may be partly explained by previously identified issues with IonTorrent technology producing higher frequencies of homopolymer sequencing errors.This leads to decreased sensitivity in INDEL calling, 39 which may bias the results.As expected, practically all APC mutations were truncating or inactivating (117/120).
Small tumors are more likely to achieve CR after treatment, seen in our previous study 13 and reported by others. 14Since patients with tumors reaching CR (either pCR or cCR) have a very low risk of recurrence and are generally not recommended adjuvant therapy even if the disease is at an advanced stage at diagnosis, we studied what determines the risk of recurrence in non-CR patients to guide whether additional treatment and more intense follow-up routines are indicated.In this analysis, a mutation in APC (or any WNT-pathway mutation) meant independently lower recurrence risks.A BRAF mutation was associated with shorter TTR in univariable survival analyses (Supplementary Figure 5C,E).These findings, if confirmed, could potentially be clinically relevant.To our surprise, we found that tumors that initially were smaller (≤5 cm) had a considerably higher recurrence risk than larger tumors (about 40% vs. 20%).We made several attempts to explain this finding (Supplementary Table 6) and found that these small tumors, remaining after treatment, more often carried a mutation in TP53 compared with large tumors.The link between a mutation in TP53 and RT-resistance reported by others 40,41 could not be found in our material, but a plausible explanation is that small non-CR tumors have a more aggressive biology with greater risk of dissemination and thus recurrence than large non-CR tumors.Furthermore, several mutational differences were seen between low, mid, and upper rectum (Supplementary Table 6).In contrast to another report, we found more APC mutations in lower rectum than in mid/upper rectum. 32 line with our previous publication, the use of clinical data only resulted in an appreciable level of predictive capability. 13The level of accuracy increased marginally (from AUC 0.76 to 0.80) with the addition of mutation status for selected genes, but this slight increase suggests that other factors may be decisive of tumor response than the ones investigated here.
The use of an unselected population-based material of rectal cancers and the retrospective review of patients' records to complete clinical data are important strengths of this study.This is one of the largest studies on treatment response prediction including both clinical and targeted sequencing data so far, but the limited number of untreated tumor tissues (biopsies) of sufficient quality and size reduced the number of patients to include in the analyses.This was partly compensated for by the use of an additional, larger, hotspot cohort in which only four genes and their hotspot mutations were considered, both to validate findings from the exploratory, targeted sequencing cohort, and to make the results more clinically relevant.
The disadvantage of this is that sequencing data has been produced with different techniques and collected from several sources.The use of FFPE material comes with issues such as crosslinking and deamination of cytosine bases that can cause misinterpretations of DNA sequences.UDG treatment and deamination score were used to mitigate FFPE issues.Another limitation is that patient-matched normal tissue-samples were not included; however, this is of little importance in the hotspot cohort.Moreover, many of the studied mutations are rare (as we have used mutational frequency of 5% as inclusion criterion) explaining why strong conclusions cannot be drawn.Multiple statistical tests have been performed in this study, increasing the risk of a chance finding.Thus, the novel findings require further investigation in other patient cohorts.Besides presenting OS, we present data on TTR rather than DFS or relapse-free survival (RFS), since TTR better reflects tumor outcome than both OS and DFS/RFS, particularly in elderly patients where deaths from other causes than the cancer are common.Another strength is that this is the first study, to our knowledge, to investigate the impact of genetic factors on recurrence risk among non-responders, the clinically relevant group for treatment after surgery and follow-up intensification.

| CONCLUSIONS
Primary tumor size and stage and treatment were strong predictors of CR.In addition, genetic factors such as SMAD4 and SYNE1 may be decisive of tumor remission and potentially used as pretreatment predictors of therapy response.Further investigations in even larger cohorts are required to establish these factors as clinically relevant.
We found that a mutation in KRAS was an independent predictor of non-CR that more than halved the odds of reaching CR, seen in both the hotspot and the targeted sequencing cohorts.Moreover, BRAF V600E meant an independently increased risk of recurrence.We also discovered that small tumors remaining after treatment were more prone to metastasize than large non-CR tumors and found mutations in TP53 as possible indicators of a more aggressive biology.These findings pave the way for personalized interventions, optimizing outcomes in the challenges of rectal cancer preoperative treatment.

2. 2 |
Patient tissues, DNA samples and molecular analyses 2.2.1 | FFPE tissue samples and targeted sequencing All tissue samples were reviewed by a pathologist (A.M., B.G.) or a trained lab technician and tumors with a tumor cell content (TCC) of ≥10% were included.DNA was extracted from formalin-fixed paraffin-embedded (FFPE) sections of pretreated biopsies using the NGEx ® FFPE DNA purification kit (Oncodia, Sweden) with the Magtration ® magLEAD 12gC automated nucleic acid extraction system.DNA concentration was measured using a broad range Qubit ® Fluorometric Quantification assay for double stranded DNA (Thermo-Fisher Scientific, USA).Uracil-DNA glycosylase (UDG) enzyme (Thermo-Fisher Scientific, USA) treatment was performed to reduce DNA sequencing artifacts introduced by FFPE processing.

F
I G U R E 1 Flow chart of study cohort selection, study groups and sequencing data sources.(A) Flow chart of study cohort selection from the total cohort of 1217 rectal cancer patients diagnosed 2010-2018 and (B) study groups and sequencing data sources.CRT, chemoradiotherapy; CT, chemotherapy; MRI, magnetic resonance imaging; scRT, short-course radiotherapy; W&W, watch-and-wait; WGS, whole-genome sequencing.
and their similarity in outcome.Overall treatment time (OTT) was the time from start of first radiotherapy until surgery or, for W&W patients, the time until the posttreatment response evaluating MRI.IBM SPSS Statistics version 28.0 and R version 4.1.3were used for analyses.Statistical significance was set to p < .05.Pearson's Chi-square test was used for group comparisons to assess associations between the chosen variable and tumor regression.Fisher's exact test was performed if >20% of cells had an expected frequency of <5.
Between 2010 and 2018, 1217 patients were diagnosed with rectal cancer in Uppsala and Dalarna regions (Figure 1A), of which 958 were non-metastatic.A total of 402 non-metastatic patients received scRT (not operated immediately or within 21 days from the first fraction), CRT or scRT+CT were included in the study cohort.Of these, 371 had delayed surgery and of the remaining 31 patients, 17 entered a W&W program after achieving cCR and 14 non-operated patients had therapy response evaluated by MRI (all non-CR).Pretreatment patient and tumor characteristics are presented in Supplementary Table 5. Treatment decision was influenced by age, TN-stage, and other MRI characteristics, with patients of younger age or advanced tumors being directed to more intensive treatment (CRT or scRT+CT).Less advanced TN-stage and older age were seen among patients treated with scRT.These patients were sometimes frail and considered not to tolerate the more intense reference treatments.The proportion of complete responders (pCR: n = 45; cCR: n = 17) increased with more intensive treatment, from 6% in the scRT group (n = 11), to 18% in the CRT group (n = 17) and 25% in the scRT+CT group (n = 34, p < .001).
U R E 2 Brick plot over the 12 most frequently mutated genes (mutational frequency ≥5%) in the targeted sequencing cohort (n = 216) in complete responders (n = 31) and non-complete responders (n = 185), separately.Each patient is represented by a vertical line while each gene, TMB, tumor remission and treatment are represented by the horizontal lines.TMB, tumor mutational burden.
of the selected predictors.In the whole cohort, treatment and tumor size demonstrated a substantial level of accuracy with an AUC of 0.76 (95% CI [0.70-0.83]).In the hotspot cohort, the addition of the CEA value and KRAS status to the aforementioned factors gave an AUC of 0.77 (95% CI [0.70-0.84]).The predictive capacity increased slightly with the use of the predictors from the targeted sequencing cohort.Here, treatment, Hb level, cT stage, and status of KRAS, SMAD4 and SYNE1 generated an AUC of 0.80 (95% CI [0.70-0.90]).
Results are given as numbers, n (%) unless indicated otherwise.Complete responders (CR) and non-complete responders (non-CR) were compared in the statistical analyses.Sex and age were not significant in any of the groups and excluded from the table.Stepwise logistic regression with backward elimination model for identification of factors associated with complete tumor remission (CR) in the different study cohorts.