Presurgical levels of circulating tumour DNA in patients with resectable chemotherapy-naïve colorectal liver metastases: association with multiorgan recurrence and survival in the MIRACLE cohort

Abstract

Background

Almost half of patients with colorectal liver metastases (CRLM) experience disease recurrence within one year after treatment with curative intent. To improve treatment, upfront identification of patients with CRLM at high risk of rapid disease recurrence is crucial.

Methods

In this retrospective cohort-study, pretreatment ctDNA levels were determined by the modified fast aneuploidy screening test-sequencing system (mFast-SeqS) in 182 patients with resectable CRLM who did not receive perioperative chemotherapy. Resulting aneuploidy scores were dichotomized using a predefined threshold and associated with clinical outcome.

Results

Of 182 analysed patients, 34 (19%) were classified as ctDNA high and 156 (81%) as ctDNA low. Recurrence-free (RFS) and overall survival (OS) were shorter in the ctDNA high versus low group, with 1-year RFS of 29% versus 52%, and 3-year OS of 48% versus 78% respectively (log-rank P = 0.029 and < 0.001 respectively). The cumulative incidence of multiorgan recurrence within the first year after local treatment was significantly higher in ctDNA-high patients (Gray’s test P < 0.001). Multivariable Cox regression analysis revealed that the aneuploidy score is independently associated with RFS (HR = 1.94; 95% c.i. 1.19 to 3.18) and multiorgan RFS up to 1 year (HR = 2.56; 95% c.i. 1.41 to 4.65) and OS up to 3 years (HR = 3.28; 95% c.i. 1.79 to 6.01).

Conclusions

This study shows that mFast-SeqS provides an affordable and minimally invasive test for the upfront recognition of patients with CRLM at increased risk of rapid (multiorgan) disease recurrence.

Our results show the impact of preoperative ctDNA levels on recurrence-free survival (RFS), especially multiorgan recurrence. High preoperative ctDNA levels are a strong predictor of a shorter RFS and overall survival after liver resection. Furthermore, the aneuploidy score acts as an independent prognostic biomarker in addition to known clinical risk factors, which could help guide the administration of preoperative chemotherapy.

Statement of surgical relevance

To improve outcome and quality of life in patients with CRLM, identifying those at high risk of rapid disease recurrence is crucial to enable better-informed treatment decisions. Biomarkers providing information before local treatment are highly relevant as they permit clinicians to tailor neoadjuvant treatment strategies and surgery. The current study demonstrated that pretreatment levels of circulating tumour DNA provide an easily implementable, minimally invasive biomarker for the upfront recognition of CRLM patients at increased risk of rapid (multiorgan) disease recurrence in the first year after local treatment. In addition, patients with high levels of ctDNA before local treatment had lower overall survival rates than patients with low ctDNA levels. This test may provide a promising tool for more personalized and better-informed treatment choices in the neoadjuvant setting in patients with resectable CRLM.

Introduction

Up to 40% of individuals with colorectal cancer (CRC) will develop colorectal liver metastases (CRLM) during the course of their disease, which significantly impacts survival in these patients^1–3. Whereas curative treatment is available for 25% of these individuals, 40% of these patients face recurrent disease within one year after surgery⁴.

The value of perioperative chemotherapy in this patient group remains debatable, as large, randomized trials showed only a relatively small effect on recurrence-free survival (RFS) and no effect on overall survival (OS)^5–7. Based on these results, local guidelines differ with respect to (neo-)adjuvant treatment schemes in these patients. In the Netherlands, chemotherapy is not standard of care for patients with CRLM suitable for local therapies. To improve patient care and oncological outcomes for individuals with CRLM, better prognostic biomarkers are needed to reduce either overtreatment or undertreatment depending on local guidelines.

An emerging novel biomarker is cell-free circulating tumour DNA (ctDNA), which can be utilized to detect minimal residual disease (MRD), monitor real-time therapy response, or to determine an individual’s prognosis^8,9. MRD and monitoring of therapy response require highly sensitive assays, as ctDNA levels following local or systemic treatment are generally low. Our previous study in CRLM patients without perioperative chemotherapy (MIRACLE cohort) demonstrated that using a CRC-specific mutation panel, ctDNA was detected in 63% of patients before treatment. Within this subgroup of patients, ctDNA positivity following local treatment was associated with a poor outcome¹⁰. To improve the detection rate and enable accurate, sensitive ctDNA detection in virtually all patients following local treatment, personalized, tumour-informed panels can be used^11,12. The potential clinical utility of ctDNA-based MRD analyses in resectable CRLM to guide adjuvant chemotherapy was demonstrated by Kataoka et al. in a recent retrospective analysis within the GALAXY study¹³.

However, as mentioned before, preferably one would be able to recognize patients in need of additional treatment before surgery to allow for neoadjuvant chemotherapy, which is associated with higher tolerability and compliance compared to adjuvant chemotherapy. Recent data suggest that at baseline ctDNA levels could be more informative than the mere presence or absence of ctDNA, as demonstrated by Reichert et al. using a cut-off of 10% ctDNA in real-world data¹⁴. Interestingly, these high levels of ctDNA do not require (ultra-)sensitive detection assays, which may facilitate future clinical implementation.

The modified fast aneuploidy screening test-sequencing system (mFast-SeqS) uses genome-wide amplification of LINE-1 elements to identify blood samples with a high circulating tumour load at low costs¹⁵. Using this assay, a genome-wide aneuploidy score of 5 or higher was previously established to be indicative of ≥10% ctDNA, which was subsequently shown to have prognostic value in advanced breast, prostate, and bladder cancer patients^16–19. The aim of the current study was to evaluate whether the mFast-SeqS assay can be used in resectable CRLM patients before surgery to identify patients at high risk of rapid (multiorgan) recurrence. For this purpose we returned to the unique MIRACLE cohort, consisting of patients with resectable CRLM who did not receive perioperative chemotherapy. In contrast to our previous work in this cohort, demonstrating an association between post-treatment presence of ctDNA and/or circulating tumour cells and clinical outcome, we here performed a thorough evaluation of the true prognostic value of high versus low pretreatment ctDNA levels without any confounding effects of perioperative chemotherapy.

Methods

Study design

The MIRACLE cohort, described in more detail elsewhere, is a prospective, observational biomarker study in patients with isolated, resectable CRLM, recruited between October 2015 and December 2021¹⁰. None of these patients received neoadjuvant chemotherapy. See Fig. 1 for the study design.

Fig. 1.

Study design and sample inclusion mFastSeqS in MIRACLE cohort

CRLM, colorectal liver metastases; IC, informed consent.

Patients with extrahepatic disease, liver-first procedures, or patients receiving chemotherapy around local treatment for the CRLM were not eligible for inclusion. The study protocol was approved by the institutional ethics committee at the initiating centre (Erasmus Medical Centre, Rotterdam, The Netherlands), and at each participating site. All participants provided written informed consent in accordance with the principles of the Declaration of Helsinki.

Pretreatment staging was done by CT scan (abdominal and thoracic), as per Dutch National Guidelines. In a minority of patients an MRI was performed, whereas PET-CT was not used in pretreatment staging. Follow-up for disease recurrence was performed according to local guidelines; every 3 months in the first 2 years after local therapy and every 6 months for the following 3 years. Follow-up consisted of clinical evaluations, serum carcinoembryonic antigen (CEA) level measurements, and liver ultrasonography and/or thoracoabdominal CT scans at least twice a year and/or when a rise in CEA occurred.

Blood sample collection and plasma isolation

Blood samples of participants were collected the evening before or on the day of local treatment (pretreatment sample). Blood was collected in CellSave tubes, which were processed within 96 h as described before²⁰. To separate plasma, blood was centrifuged at 1700×g for 10 min. Subsequently, this plasma was transferred to a new tube and a second centrifugation step was performed at 12 000×g for 10 min at 4°C. Plasma was then stored at −80°C until further use.

Cell-free DNA isolation

For cell-free DNA (cfDNA) extraction, 4 ml of plasma was processed using the QIAamp Circulating Nucleic Acid kit (Qiagen) according to manufacturer’s instructions. Isolated cfDNA was quantified by the Quanti-IT dsDNA High-sensitivity Assay (Invitrogen) as per manufacturer’s protocol and the readout was generated by the Qubit Fluorometer (Invitrogen). Afterwards, cfDNA was stored at −20°C.

mFast-SeqS method

Aneuploidy scores were generated by the mFast-SeqS method as described previously using 1 ng of cfDNA¹⁶. In short, 1 ng of cfDNA was amplified using a Phusion high-fidelity polymerase (New England Biolabs) by a single primary primer pair for specific amplification of LINE-1 sequences throughout the genome (2′ at 98 °C, followed by 8 cycles of 10′ at 98 °C, 2′ at 57 °C and 2′ at 72 °C). To increase the complexity of the resulting sequencing libraries, a random spacer was introduced to the forward primer as described by Fadrosh et al.²¹.

In the second PCR step, Illumina indexes and adapters were added using the following PCR programme: 2 min at 98°C, followed by 18 cycles of 10 s at 98°C, 15 s at 65°C, and 15 s at 72°C. Resulting libraries were pooled equimolarly and sequenced on a MiSeq system, generating at least 90 000 150 bp single-end reads. Read counts per chromosome arm were normalized to the total library size, and subsequently, a Z-score per chromosome arm was calculated relative to healthy controls (17 males and 18 females), excluding the short arms of chromosomes 13, 14, 15, 21, and 22, as well as chromosomes X and Y. Finally, the resulting Z-scores were squared and summed and compared to squared and summed values of the controls, yielding a genome-wide aneuploidy score. This aneuploidy score indicates the number of standard deviations a sample differs from the healthy controls. An aneuploidy score of ≥5 was predefined as ctDNA high, as this cut-off value has been shown to select plasma samples with a true tumour-derived aneuploidy signal¹⁵. This cut-off corresponds to a ctDNA fraction of ≥10% and was previously shown to be an independent prognostic factor in advanced cancer^14,16–19.

Statistics

Baseline characteristics of patients with detectable ctDNA versus no detectable ctDNA at baseline were compared using the Kruskal–Wallis test for continuous variables and the Fischer’s exact test for discrete variables. Continuous variables are represented as median with their corresponding 95% confidence intervals. RFS was defined as the time from liver resection or ablation until diagnosis of locoregional or distant recurrence on radiographical imaging or until death or last follow-up. OS was defined from the date of treatment until the date of last follow-up or death. The primary endpoints were 1-year RFS, 1-year multiorgan recurrence, and 3-year OS.

The primary endpoints were chosen to reflect clinically meaningful events within relevant time frames: 1-year RFS captures the immediate risk window (early recurrences), 1-year multiorgan recurrence provides insight into the extent and aggressiveness of disease dissemination, and 3-year OS offers a longer-term perspective on survival in this patient cohort. One-year RFS and 3-year OS were assessed through the Kaplan–Meier method and compared using the log-rank test. For 1-year multiorgan recurrence, the cumulative incidence function (CIF) was estimated using the Fine and Gray competing risks model. Differences between groups were assessed using Gray’s test²². In this analysis, the competing risk was defined as the diagnosis of solitary-organ recurrence treated with systemic chemotherapy within 1 year, as such treatment is expected to modify the subsequent risk of multi-organ recurrence during that period. Uni- and multivariable Fine and Gray hazard models were used to assess associations with multiorgan recurrence up to 1 year.

Uni- and multivariable Cox proportional hazard was performed to investigate the Z-score (dichotomized aneuploidy score) as a potential predictor in terms of long-term oncological outcomes. Established clinical risk factors were assessed using univariable analysis, and those with P < 0.10 were subsequently included as covariates in the multivariable analysis for adjustment. Results were reported as hazard ratios with 95% confidence intervals. In multivariable analysis, P < 0.05 was considered statistically significant. All statistical analyses were performed using R software version 1.4 (R project for Statistical Computing).

Results

Baseline characteristics

Baseline samples were available for 188 patients, of which 182 were successfully analysed by mFast-SeqS. Six samples were excluded, as the remaining cfDNA content was insufficient for analysis, see Fig. 1 for a detailed overview of patient and sample inclusion. Using the Z-score/dichotomized aneuploidy score, 34 patients were classified as ctDNA high (Z-score ≥ 5, 19% of patients) and 148 as ctDNA low (Z-score < 5, Fig. 2). Baseline characteristics are shown in Table 1 separately for ctDNA high and ctDNA low patients. Patients with high ctDNA levels more often were female (P = 0.007) and had a hepatic metastasis with a diameter above 5 cm (P < 0.001). Further investigations showed that, compared to male patients, female patients in our cohort more often had metastatic lesions above 5 cm (4% versus 16%, chi-square P = 0.005), indicating a higher disease burden in female patients. The association between disease burden and ctDNA levels also becomes apparent in the distribution of parenchyma-sparing resection versus major resection over ctDNA-low and ctDNA-high patients. Median follow-up was 45 months (i.q.r. 30–62 months).

Fig. 2.

Detection rates of chromosomal aneuploidy

Table 1.

Baseline characteristics

	Baseline
	ctDNA-low	ctDNA-high	P
N	148	34
sex, N (%)
Female	39 (26.4)	17 (50.0)	0.007*
Male	109 (73.6)	17 (50.0)
Age at resection of CRLM (median (i.q.r.))	66.5 (61.0, 74.0)	66.5 (61.0, 73.5)	0.881
ASA class, N (%)
1–2	104 (70.3)	22 (64.7)	0.526
3–4	44 (29.7)	12 (35.3)
Location of primary tumour, N (%)
Left-sided	65 (42.6)	15 (44.1)	0.298
Rectum	46 (31.1)	14 (41.2)
Right-sided	39 (26.4)	5 (14.7)
T stage of primary tumour, N (%)
1–2	30 (20.3)	3 (8.8)	0.191
3–4	115 (77.7)	31 (91.2)
Unknown	3 (2.0)	0 (0.0)
Lymph nodes of primary tumour, N (%)
N0	57 (38.5)	10 (29.4)	0.393
N+	88 (59.5)	24 (70.6)
Unknown	3 (2.0)	0 (0.0)
Metachronous/synchronous, N (%)
Metachronous	91 (61.5)	25 (73.5)	0.188
Synchronous	57 (38.5)	9 (26.5)
Disease-free interval primary—CRLM, N (%)
≤1 year	102 (68.9)	20 (58.8)	0.259
>1 year	46 (31.1)	14 (41.2)
Number of CRLM, N (%)
1	81 (54.7)	14 (41.2)	0.154
>1	67 (45.3)	20 (58.8)
Diameter of largest CRLM, N (%)
≤5	144 (97.3)	24 (70.6)	<0.001*
>5	3 (2.0)	10 (29.4)
Unknown	1 (0.7)	0 (0.0)
Elevated CEA before resection of CRLM, N (%)
<5	38 (25.7)	8 (23.5)	0.896
>5	85 (57.4)	21 (61.8)
Unknown	25 (16.9)	5 (14.7)
KRAS status (tissue), N (%)
Mutated	28 (18.9)	5 (14.7)	0.578
Wildtype	28 (18.9)	9 (26.5)
Unknown	92 (62.2)	20 (58.8)
NRAS status (tissue), N (%)
Mutated	0 (0.0)	0 (0.0)	0.355
Wildtype	44 (29.7)	13 (38.2)
Unknown	104 (70.3)	21 (61.8)
BRAF status (tissue), N (%)
Mutated	1 (0.7)	1 (2.9)	0.413
Wildtype	53 (35.8)	14 (41.2)
Unknown	94 (63.5)	19 (55.9)
MSI status (tissue), N (%)
MSI	3 (2.0)	0 (0.0)	0.420
MSS	80 (54.1)	22 (64.7)
Unknown	65 (43.9)	12 (35.3)
KRAS status (cfDNA), N (%)†
Mutated	38 (25.9)	13 (38.2)	0.311
Wildtype	109 (73.6)	21 (61.8)
Unknown	1 (0.7)	0 (0.0)
NRAS status (cfDNA), N (%)†
Mutated	5 (3.4)	0 (0.0)	0.490
Wildtype	142 (95.9)	34 (100)
Unknown	1 (0.7)	0 (0.0)
BRAF status (cfDNA), N (%)†
Mutated	2 (1.4)	1 (2.9)	0.720
Wildtype	145 (97.9)	33 (97.1)
Unknown	1 (0.7)	0 (0.0)
Risk group according to Fong et al.23, N (%)
High	36 (24.3)	13 (38.3)	0.188
Low	110 (74.3)	20 (58.8)
Unknown	2 (1.4)	1 (2.9)
Resection type, N (%)
Primary first	129 (87.2)	32 (94.1)	0.503
Synchronous	18 (12.2)	2 (5.9)
Unknown	1 (0.7)	0 (0.0)
Resection margin of CRLM, N (%)
R0	134 (90.5)	31 (91.2)	0.384
R1	8 (5.4)	3 (8.8)
Unknown	6 (4.1)	0 (0.0)
Parenchym-sparing resection, N (%)
Yes	114 (77.0)	20 (58.8)	0.030*
No	34 (23.0)	14 (41.2)
Major resection, N (%)
Yes	14 (9.5)	11 (32.4)	<0.001*
No	134 (90.5)	23 (67.7)
RFA/MWA
Yes	42 (28.4)	8 (23.5)	0.747
No	105 (70.9)	26 (76.5)
Unknown	1 (0.7)	0 (0.0)

^*Significant result. ctDNA-low is defined as an aneuploidy score <5, ctDNA-high as an aneuploidy score ≥5. †Mutation status determined by the Oncomine colon cfDNA Assay (Thermofischer). CLRM, colorectal liver metastases; i.q.r., interquartile range; ASA, American Society of Anesthesiologists; CEA, carcinoembryonic antigen; RFA, radiofrequency ablation; MWA, microwave ablation; MSI, microsatellite instability.

Oncological outcomes and recurrence patterns

RFS and OS were shorter in the ctDNA high versus low group, with 1-year RFS of 29% versus 52%, and 3-year OS of 48% versus 78% respectively (log-rank P = 0.029 and <0.001 respectively; Fig. 3 and Fig. S1). The median RFS of ctDNA high patients versus ctDNA low patients was 7 months (95% c.i. 6 to 12) and 13 months (95% c.i. 10 to 18) respectively. Median OS time of patients in the ctDNA high group was 36 months (95% c.i. 28—not reached. The median OS time for patients in the ctDNA low group was not reached. As we found ctDNA-high patients were more often female and had metastatic lesions above 5 cm, we also compared RFS and OS between ctDNA-high and ctDNA-low patients separately for males, females, and patients with metastatic lesions up to 5 cm. All subgroup analyses showed ctDNA-high patients had significantly worse OS compared to ctDNA-low patients, confirming the independent prognostic value of pretreatment ctDNA levels (Figs S2, S3).

Fig. 3.

Recurrence-free and overall survival rates stratified by Z-score (ctDNA high versus ctDNA low)

OS, overall survival.

During follow-up, 28/34 (82%) patients with ctDNA-high and 101/148 (68%) in the ctDNA-low group were faced with recurrence (P = 0.10). Disease recurrence was found to affect multiple organs in 59% of patients in the ctDNA-high group, compared to 24% of patients in the ctDNA-low group (P < 0.001). Multiorgan recurrences within one year after treatment were diagnosed in 44% of patients in the ctDNA-high group and 13% of patients in the ctDNA-low group (P < 0.001). The cumulative incidence of multiorgan recurrence within the first year after local treatment was significantly higher in patients with detectable ctDNA (Gray’s test P < 0.001, Fig. 4 and Fig. S4). With respect to recurrence patterns, 50% of ctDNA-high patients showed extrahepatic recurrence within one year after surgery compared to 23% of ctDNA-low patients (P = 0.004). The remaining 77% of ctDNA-low patients either had no recurrence or liver-only disease, potentially salvageable by repeated local treatment. Only 29% in the ctDNA-high group had liver-limited recurrence, as opposed to 47% in the ctDNA-low group (P = 0.12). At time of recurrence (n = 129), palliative care was recommended for 53 patients and additional treatment with curative intent for 59 patients. Data about curative or palliative treatment at time of recurrence was unavailable for 17 patients. A significantly higher proportion of patients in the ctDNA high group were referred to palliative care once they recurred compared to patients in the ctDNA low group (61% versus 35%, chi square P = 0.054).

Fig. 4.

For 1-year multiorgan recurrence, the cumulative incidence function (CIF) was estimated using the Fine and Gray competing risks model to account for the occurrence of single-organ recurrence as competing event

The graph shows the cumulative incidence of multiorgan recurrence stratified by Z-score (ctDNA high versus ctDNA low).

Uni- and multivariable proportional hazard analyses

To evaluate whether the observed prognostic value of the Z-score (high versus low) is independent from already known prognostic clinical parameters, Cox proportional hazards regression models were performed for RFS and multiorgan RFS up to 1 year and OS up to 3 years, testing all relevant clinical characteristics (Table 2) and the Z-score. Upon univariable analysis using a Cox proportional hazard model, the Z-score was a significant predictor for RFS up to 1 year (Table 2). Similarly, rectum as location of the primary tumour, positive lymph nodes of the primary tumour, a shorter disease-free interval between resection of the primary tumour and detection of CRLM, and a higher number of CRLM were all predictors for a shorter RFS up to 1 year. Multivariable analysis corrected for the significant clinical variables in the univariable analysis revealed an HR of 1.94 (95% c.i.: 1.19 to 3.18) for patients with a high level of ctDNA (Z-score ≥ 5).

Table 2.

Univariable and multivariable Cox proportional hazard model for 1-year recurrence-free survival. Univariable and multivariable Fine and Gray model for 1-year multiorgan recurrence-free survival

Variable		Univariable 1-year RFS HR (95% c.i.)	Multivariable 1-year RFS HR (95% c.i.)	Univariable 1-year RFS (multiorgan) HR (95% c.i.)	Multivariable 1-year RFS (multiorgan) HR (95% c.i.)
Age at resection (cont.)		0.99 (0.97–1.01)		1.02 (0.99–1.06)
Gender (male)		0.88 (0.57–1.36)		0.66 (0.36–1.22)
Location of primary tumour
Left-sided (ref)
Right-sided		1.12 (0.69–1.82)	1.04 (0.63–1.70)	1.18 (0.55–2.52)
Rectum		0.55 (0.33–0.91)*	0.56 (0.33–0.95)*	1.18 (0.59–2.52)
Lymph nodes of primary tumour (N+)		1.70 (1.09–2.64)*	1.53 (0.98–2.40)	1.67 (0.83–3.36)
DFI (cont.)		0.96 (0.94–0.98)‡	0.96 (0.94–0.99)‡	0.98 (0.95–1.01)
Number of CRLM (cont.)		1.26 (1.13–1.40)‡	1.23 (1.19–1.38)‡	1.27 (1.11–1.45)‡	1.19 (1.04–1.36)†
Diameter of largest CRLM (cont.)		1.03 (0.93–1.13)		1.03 (0.91–1.16)
Preoperative CEA (cont.)		1.00 (1.00–1.00)		1.00 (1.00–1.00)
Baseline Z score (ctDNA high)		1.64 (1.03–2.62)*	1.94 (1.19–3.18)†	2.97 (1.68–5.24)‡	2.56 (1.41–4.65)†

Disease-free interval (DFI) between primary tumour and detection of colorectal liver metastasis (CRLM).

^* P < 0.05. †P < 0.01. ‡P < 0.001. RFS, recurrence-free survival; CEA, carcinoembryonic antigen.

A Fine and Gray proportional hazard analysis was performed for multiorgan recurrence up to 1 year after local treatment. Univariable analysis revealed a high Z-score and a higher number of CRLM as predictors for multiorgan recurrence. When corrected for each other, both variables remained significant predictors in multivariable analysis with an HR of 2.97 (95% c.i.: 1.68 to 5.24) for patients with a high level of ctDNA (Z-score ≥ 5).

Upon univariable analysis, a Z-score of ≥5, positive lymph nodes in the primary tumour, and a shorter disease-free interval between the primary tumour and detection of the CRLM were associated with decreased OS rates up to 3 years (Table 3). Subsequent multivariable analysis including the significant features from the univariable analysis revealed the independent value of the Z-score for OS up to 3 years in patients with resectable CRLM (HR 3.28; 95% c.i.: 1.79 to 6.01).

Table 3.

Univariable and multivariable Cox proportional hazard model for 3-year overall survival

Variable		Univariable 3-year OS HR (95% c.i.)	Multivariable 3-year OS HR (95% c.i.)
Age at resection (cont.)		1.03 (0.99–1.06)
Gender (male)		0.77 (0.43–1.39)
Location of primary tumour
Left-sided (ref)
Right-sided		1.09 (0.54–2.22)
Rectum		0.95 (0.49–1.84)
Lymph nodes of primary tumour (N+)		1.91 (0.99–3.70)	1.69 (0.87–3.28)
DFI (cont.)		0.98 (0.95–1.00)	0.98 (0.95–1.00)
Number of CRLM (cont.)		1.13 (0.97–1.33)
Diameter of largest CRLM (cont.)		0.95 (0.79–1.15)
Preoperative CEA (cont.)		1.00 (0.99–1.00)
Baseline Z score (ctDNA high)		2.86 (1.58–5.17)‡	3.28 (1.79–6.01)‡

Disease-free interval (DFI) between primary tumour and detection of colorectal liver metastasis (CRLM).

^‡ P < 0.001. OS, overall survival; CEA, carcinoembryonic antigen.

Discussion

This study shows the prognostic value of mFast-SeqS aneuploidy scores in the setting of resectable, chemotherapy-naïve CRLM. Patients with high levels of ctDNA before local treatment had lower OS rates than patients with low ctDNA levels. Additionally, around half of patients with ctDNA-high levels before local treatment developed multiorgan recurrence within one year following surgery, suggesting the need for systemic treatment. This straightforward, affordable, and minimally invasive test may provide a promising tool for more personalized and better-informed treatment choices in the pretreatment setting in patients with resectable CRLM in order to increase overall survival rates as well as quality of life.

Although patients with resectable CRLM undergo treatment with curative intent, many are faced with disease recurrence within the first year after surgery⁴. Additional perioperative chemotherapy was shown to increase disease-free but not overall survival, suggesting only a subgroup of patients derives benefit from this additional treatment^5–7. This was supported by the recent finding that survival benefit from chemotherapy was only observed in CRLM patients with remaining ctDNA following surgery¹³. In this study, we evaluated whether pretreatment levels of ctDNA in patients with resectable CRLM could recognize patients with a poor prognosis who may benefit from additional systemic treatment.

An impaired 3-year OS and 1-year RFS was found in patients with high ctDNA levels before local treatment. In particular, patients with high ctDNA levels more often showed multiorgan recurrence compared to patients with low baseline ctDNA levels. Results remained significant in multivariable analyses corrected for known clinical risk factors. As ctDNA levels reflect tumour load as well as tumour proliferation rate, this may explain the observed associations with recurrence patterns and OS rates^24,25. In this cohort, we observed relatively more females with high levels of ctDNA compared to males, suggesting a higher tumour load in these patients. Several other studies have described the same phenomenon, reporting worse survival outcomes in females. A clear explanation remains to be found, but both biological (for example the immune system) and non-biological (for example local versus systemic treatment) factors are hypothesized to play a role^26–28.

Previous studies were not able to show a convincing association between preoperative ctDNA levels and OS^13,29. This may be explained by the differences in detection methods. Two small studies investigating a specific alteration in KRAS in limited sample sizes suggested an association between ctDNA presence and OS^30,31. However, when almost all patients have detectable ctDNA at baseline, this may not be as informative in risk stratification. For example, the CIRCULATE-Japan GALAXY study reported a preoperative detection of ctDNA in 98% of patients in their cohort, but no association between the presence of preoperative ctDNA and long-term survival¹³. On the other hand, an increasing number of studies demonstrate a clear prognostic value for pretreatment ctDNA levels in various cohorts of advanced cancer patients^14,16,17. The current study therefore evaluated whether baseline ctDNA levels, as opposed to ctDNA presence, may provide prognostic information in patients with resectable CRLM by dividing patients in ctDNA high and low.

In order to add to this existing body of evidence, this study investigated post-treatment recurrence patterns in relation to ctDNA levels before local treatment, whereupon neoadjuvant chemotherapy may still be administered. In our cohort, high pretreatment ctDNA levels were particularly associated with rapid multiorgan recurrence, suggesting more aggressive disease and perhaps already existing occult disease at time of treatment.

These results suggest preoperative ctDNA levels could potentially be used to recognize a subgroup of patients with resectable CRLM at high risk of multiorgan recurrence. As recent data suggest that patients with multiorgan metastatic colorectal cancer do not benefit from local treatment but require systemic treatment, this may reflect an important additional outcome to consider from a clinical perspective³². In contrast, patients with low ctDNA levels before surgery less frequently recurred within the first year after surgery and more often developed recurrences limited to the liver or lung. Additional local treatment modalities are frequently available for these patients, suggesting they may be spared the adverse effects associated with chemotherapy.

Pretreatment risk stratification, as opposed to post-treatment, has several advantages in the medical decision-making process. Neoadjuvant chemotherapy can simplify surgical procedures by tumour down-sizing and has a higher tolerability and compliance compared to adjuvant chemotherapy. In support of this, the JCOG0603 trial reported severe adverse events in approximately half of the patients receiving adjuvant chemotherapy⁶. On the other hand, neoadjuvant chemotherapy is also associated with potential drawbacks, including chemotherapy-induced liver injury and the risk of disease progression during systemic treatment, which may compromise liver resection in a small amount of patients^5,33,34. Taken together, this further underlines the need for upfront risk stratification based on both clinical parameters and ctDNA measurements to better delineate which therapeutic strategies confer the greatest benefit across these risk groups. For example, one could envision that patients with favourable prognostic factors and low preoperative ctDNA levels may be cured by local treatment(s) only, whereas patients at high risk of rapid (multiorgan) recurrence may in turn be the group that does benefit from additional neoadjuvant chemotherapy¹³. Conversely, neoadjuvant chemotherapy could also be considered for selected low-risk patients to further improve curative potential. The patient group with a very poor prognosis may be better served by omitting local treatment all together to focus on quality of life and overall welfare in the last stages of life. Either way, to increase chances of survival in patients with resectable CRLM while maintaining quality of life, it is crucial to evolve towards more judicious patient selection for chemotherapy and, thus, more personalized treatment strategies.

Advantages of the current study include the homogeneous cohort, in which none of the patients received perioperative chemotherapy, and the use of an easily implementable tumour-agnostic assay to measure ctDNA levels with a predefined cut-off. The mFast-SeqS only requires 1 ng cfDNA, has a fast turnaround time, and only requires ultralow sequencing (0.1 M reads). Besides to being more affordable, it does not require special equipment or complex bioinformatical pipelines. In addition, it relies on the detection of chromosomal aneuploidy, one of the hallmarks of solid cancers³⁵. This eliminates the need for prior knowledge on the tumour’s genetic make-up and renders this assay applicable to virtually all patients suffering from advanced solid cancers^16–19. On the other hand, this assay does not provide information on the mutational status of genes important for clinical decision-making in metastatic colorectal cancer.

Other limitations to the current study include the fact that the cut-off used to discriminate ctDNA-high from ctDNA-low patients was predefined based on previous results in other patient cohorts, and may not have been the most optimal cut-off for this specific clinical question^15–17. However, our choice for the use of a predefined cut-off eliminates the risk for overfitting and increases reproducibility. Future work could further investigate the optimal cut-off as well as multivariable models including both ctDNA levels and relevant clinical parameters. Additionally, results reported here concern an exploratory analysis using the MIRACLE cohort, which was not specifically powered for these analyses and thus requires independent validation¹⁰. However, these first results in a substantial cohort show promising evidence of its association with multiorgan recurrence and overall survival in patients with resectable CRLM.

Funding

This study was funded by the Dutch Cancer Society (KWF: EMCR 2014-6340). The funder had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Author contributions

Lissa Wullaert (Data curation, Formal analysis, Investigation, Project administration, Resources, Software, Visualization, Writing—original draft, Writing—review & editing), Lotte van Leeuwen (Data curation, Formal analysis, Investigation, Project administration, Resources, Software, Visualization, Writing—original draft, Writing—review & editing), Vanja de Weerd (Investigation, Resources, Writing—review & editing), Mai Van (Investigation, Resources, Writing—review & editing), Esther Oomen-De Hoop (Formal analysis, Investigation, Resources, Software, Writing—review & editing), Jaco Kraan (Investigation, Resources, Writing—review & editing), Maurice Jansen (Investigation, Resources, Writing—review & editing), John W.M. Martens (Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Supervision, Validation, Writing—review & editing), Dirk Grünhagen (Conceptualization, Data curation, Investigation, Methodology, Resources, Supervision, Visualization, Writing—review & editing), Henk Verheul (Conceptualization, Funding acquisition, Investigation, Methodology, Resources, Supervision, Validation, Writing—review & editing), Cornelis Verhoef (Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Writing—review & editing), and Saskia Wilting (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing)

Disclosure

Institutional research grant support from Pfizer (S.M.W., J.W.M.M.), Menarini (J.W.M.M.), GSK (J.W.M.M.), Tzu genomics (J.W.M.M.), and Bayer (J.W.M.M.).

Supplementary material

Supplementary material is available at BJS online.

Data availability

All data underlying the results are available upon request.