Introduction

Histopathological growth patterns (HGPs) describe distinct phenotypes of tumour growth at the transition zone between pre-existing liver parenchyma and colorectal liver metastases (CRLM) [1]. HGPs have been associated with prognosis in patients undergoing resection of CRLM [1-9]. The determination of HGPs has been standardised in international guidelines [1]. Three main HGP phenotypes are recognised: the replacement, the pushing, and the desmoplastic type HGP (Figure 1) [1, 10]. Based on prognosis, a dichotomy can be made. Patients with any observed non-desmoplastic HGP (i.e. any pushing or replacement HGP) have worse survival outcomes compared to patients with pure desmoplastic HGP [5]. This difference in survival was less apparent for patients treated with preoperative systemic chemotherapy [5]. Furthermore, higher proportions of the desmoplastic HGP were observed in pre-treated patients. These results suggest that preoperative chemotherapy may affect the HGP and raises questions regarding the assessment and value of this biomarker after preoperative systemic treatment. These results require external validation.

Examples of the distinct HGPs. (A) Example of replacement type HGP in which tumour cells ‘replace’ hepatocytes and infiltrate the liver parenchyma with direct tumour–liver cell contact. (B) Example of pushing type HGP in which the liver parenchyma is ‘pushed’ aside but is not infiltrated. No direct tumour–liver cell contact is present. (C) Example of desmoplastic type HGP, in which the tumour is separated from the liver parenchyma by a desmoplastic capsule. No direct tumour–liver cell contact is present.

The European Organization for Research and Treatment of Cancer (EORTC) intergroup study 40983 randomised controlled phase III trial compared surgery alone to surgery combined with perioperative systemic chemotherapy in patients with resectable CRLM [11, 12].

This study evaluates the effect of preoperative systemic chemotherapy on the HGPs of CRLM in an original cohort of consecutive patients undergoing resection in the Netherlands, a similar external validation cohort of patients treated in the USA, and in a post hoc analysis of a subset from the EORTC 40983 randomised controlled clinical trial.

Materials and methods

The current study was performed according to the STROBE guidelines for cohort studies and approved by the medical ethics committee of the Erasmus University Medical Centre Rotterdam (MEC 2018-1743) [13]. A waiver for renewed written informed consent was granted.

Original cohort

All consecutive patients undergoing first resection of CRLM between January 2000 and February 2019 at the Erasmus MC Cancer Institute (Rotterdam, The Netherlands) were evaluated for eligibility. Part of this cohort was previously described by Galjart et al [5]. In accordance with the previous study, patients with incomplete resection, treated by ablation only, or in whom the HGP could not be determined were excluded. Patient characteristics, primary tumour and CRLM characteristics, treatment details, follow-up, and disease recurrence were extracted from a prospectively maintained database.

External validation cohort

All consecutive patients undergoing first resection of CRLM between January 2000 and January 2019 at the Memorial Sloan Kettering Cancer Center (MSKCC) (New York City, NY, USA) were considered for inclusion in the external validation cohort. Similar exclusion criteria were applied. In addition, patients receiving preoperative hepatic arterial infusion pump (HAIP) chemotherapy were excluded as this study evaluates the relationship between HGPs and preoperative systemic chemotherapy. Data regarding patient characteristics, primary tumour and CRLM characteristics, treatment details, follow-up, and disease recurrence were also extracted from a prospectively maintained database.

Randomised patient cohort

A subset of patients from the EORTC 40983 trial (NCT00006479), from whom digitalised haematoxylin and eosin (H&E)-stained tissue sections were available, were included for post hoc analysis. This subset of patients has been described previously [14]. The details of the original trial including its short- and long-term results are reported elsewhere [11, 12]. In summary, the EORTC 40983 trial randomised 364 patients with up to four resectable CRLM between either perioperative chemotherapy and resection (CTx arm) or resection only (Rx arm). Perioperative chemotherapy consisted of the FOLFOX4 regimen with six planned preoperative and six planned postoperative cycles [15].

HGP determination

Determination of HGPs was done in accordance with international consensus guidelines [1]. Assessment was performed by light microscopy on all available H&E-stained tissue sections from all resected CRLM and blinded for outcome, preoperative treatment status, and all other clinicopathological patient characteristics. Assessment was performed by trained observers (PMHN, DJH, EPvdS, and BG) together/in consultation with a dedicated HGP pathologist (PBV) [1]. For the EORTC 40983 trial, patient assessment was performed on digitalised H&E-stained tissue sections [14] by trained observers (PMHN, DJH, and BG) and a dedicated HGP pathologist (PBV) separately. Discordant cases were subsequently reviewed by all observers together (PMHN, DJH, BG, and PBV) to achieve consensus. As multiple HGPs can be present in a single tumour, the entire tumour–liver interface on each slide was examined. During assessment, the relative fraction of the total length of the interface of desmoplastic, replacement, and/or pushing HGP was estimated and expressed as percentage. Herein, each proportion of the interface representing 5% or more was taken into account. Metastasis level estimates were calculated with equal weights assigned to individual tissue sections. The final patient level HGP scores were subsequently calculated with equal weights assigned to individual metastases. The average presence of each distinct HGP observed was determined in each of the three cohorts and stratified for preoperative treatment status. The proportional distribution of distinct HGPs was displayed graphically and stratified for preoperative treatment status, in which the horizontal axis represented individual patients and the vertical axis the corresponding observed proportion of each distinct HGP at the tumour–liver interface. The average presence of each distinct HGP was represented by its surface area. HGP determination was not performed if no viable tumour was present, in cases with inadequate tissue preservation of H&E-stained tissue section(s), or if less than 20% of the tumour–liver interface was assessable [1]. In accordance with previous findings, patients were classified as either pure desmoplastic HGP (i.e. 100% desmoplastic HGP) or non-desmoplastic HGP (any replacement and/or pushing HGP) [5, 16]. A simplified decision tree to determine the HGP on a patient level based on this clinically relevant distinction, adapted with permission from van Dam et al [1], is provided in supplementary material, Figure S1. With regard to preoperative treatment stratification, patients in the original cohort and the external validation cohort who received any systemic chemotherapy within 6 months prior to CRLM resection – with the exception of capecitabine as radiosensitiser in the treatment for rectal cancer – were considered preoperatively treated. In addition, several examples of the desmoplastic HGP with and without preoperative systemic chemotherapy were selected and were evaluated in a descriptive manner.

Tumour regression grading

For the subset of the EORTC 40983 trial, three separate tumour regression gradings were available: the Mandard tumour regression grade (TRG) [17], the mean percentage of tumour cells according to Blazer et al [18], and the histological tumour regression according to Rubbia-Brandt [19]. These three tumour regression gradings were all determined prior to the conception of this study, by an independent senior pathologist (CJ) not involved in HGP assessment and blinded for treatment arm and patient outcome. The Mandard TRG recognises five grades: 1 – absence of cancer cells replaced by abundant fibrosis; 2 – rare residual cancer cells scattered throughout abundant fibrosis; 3 – increase in the number of cancer cells but fibrosis remains predominant; 4 – residual cancer outgrowing fibrosis; and 5 – absence of regressive changes [17]. The method described by Blazer et al assesses pathological response to preoperative chemotherapy in patients with CRLM by semi-quantitatively estimating the percentage of viable tumour in relation to tumour surface area [18]. The histological tumour regression according to Rubbia-Brandt is an adaptation of the Mandard TRG and recognises three grades of tumour regression in CRLM: no histological tumour regressive or response changes (NHR), partial histological tumour response (PHR), and major or complete histological tumour response (MjHR) [19]. Tumour regression according to all three grading systems was correlated with HGP stratified for treatment arm.

Statistical analysis

Categorical data are reported using absolute numbers and corresponding percentages and continuous data using medians with corresponding interquartile ranges (IQR). Proportional differences were evaluated with the chi-squared test. Differences in medians between two groups were assessed using the Mann–Whitney U-test. The average presence of distinct HGPs was compared across preoperative treatment status by means of a parametric t-test. To evaluate whether preoperative chemotherapy was associated with the observed proportion of the desmoplastic HGP at the interface, uni- and multi-variable linear regression analyses were performed and expressed using the β coefficient with corresponding 95% confidence intervals (CIs). Additional uni- and multi-variable linear regression models were computed in a combined cohort of all patients with available data on APC, KRAS, NRAS, and BRAF mutational status, as well as on microsatellite instability (MSI) status. The association between tumour regression and the desmoplastic HGP was assessed in the trial cohort for each of the three gradings and in each treatment arm separately by multivariable logistic regression. Results are graphically displayed using scatter plots with corresponding regression line and are reported using the β coefficient with corresponding 95% CI. The reversed Kaplan–Meier method was applied to estimate the median follow-up time for survivors. Overall survival (OS) was defined as the time in months from the date of resection until the date of death. When alive, patients were censored at the date of last follow-up. Kaplan–Meier analysis was used to determine survival estimates which were compared by means of the log-rank test. Uni- and multi-variable Cox regression analyses for OS were performed in the original and the external validation cohort to correct for potential confounding. In these cohorts, survival analyses on the HGP stratified by preoperative chemotherapy have previously been performed and were therefore not repeated [5, 16, 20]. Results of the Cox regression analyses were expressed using hazard ratios (HRs) and corresponding 95% CIs. For the EORTC 40983 trial subset, the OS difference between treatment arms was estimated and compared to the long-term results of the entire trial (expressed as HR with corresponding 95% CI) [12]. In an attempt to assess differences between pre-treated and chemo-naïve patients with a desmoplastic HGP (i.e. 100% desmoplastic), clinicopathological factors and OS were compared between these subgroups in a combined cohort of all available patients. All analyses were performed using R version 4.1.0 (http://www.r-project.org).

Results Original cohort

At the Erasmus MC Cancer Institute, 1,257 patients were treated surgically for CRLM between January 2000 and February 2019. Patients were excluded due to incomplete resection of CRLM (n = 133), ablative therapy only (n = 33), and unsuitable or unavailable H&E-stained tissue sections for HGP determination (n = 214). The remaining 877 (70%) patients were included for analysis. Preoperative systemic chemotherapy was administered to 462 patients (53%). Baseline patient characteristics stratified by preoperative treatment are presented in Table 1. A graphical display of the distinct HGPs stratified for preoperative treatment status is shown in Figure 2. The average presence of desmoplastic HGP observed at the interface was 43% in chemo-naïve versus 67% in preoperatively treated patients (p < 0.001; Figure 2D). Preoperative systemic chemotherapy was independently associated with a higher proportion of desmoplastic HGP observed (adjusted β [95% CI]: 24.57 [18.28–30.87], p < 0.001; Table 2). On multivariable analysis, a non-desmoplastic HGP was associated with an adjusted HR (95% CI) for OS of 1.56 (1.23–1.98) (p < 0.001; see supplementary material, Table S1) [5, 16].

Table 1. Baseline characteristics of all three cohorts stratified by preoperative treatment status. Original cohort Erasmus MC Cancer Institute External validation cohort MSKCC Randomised patient cohort EORTC 40983 trial Preoperative chemotherapy Preoperative chemotherapy Treatment arm No Yes No Yes Rx arm CTx arm n = 462 (%) n = 415 (%) P value* n = 410 (%) n = 793 (%) P value* n = 40 (%) n = 30 (%) P value* Age at resection CRLM (median [IQR]) 66.0 [59.2–73.0] 63.0 [56.0–69.0] <0.001 62.0 [52.0–72.0] 57.0 [48.0–66.0] <0.001 67.5 [59.8–72.0] 65.0 [58.5–71.8] 0.536 Gender Male 298 (65) 270 (65) 0.863 228 (56) 452 (57) 0.645 22 (55) 19 (63) 0.484 Female 164 (35) 145 (35) 182 (44) 341 (43) 18 (45) 11 (37) Primary tumour location Right-sided 80 (18) 65 (16) 0.655 141 (36) 204 (27) 0.003 8 (20) 8 (27) 0.768 Left-sided 195 (43) 175 (43) 175 (44) 331 (45) 15 (38) 11 (37) Rectal 175 (39) 170 (41) 80 (20) 207 (28) 16 (40) 11 (37) Missing 12 (3) 5 (1) 14 (3) 51 (6) 1† (2) 0 (0) Adjuvant CTx for primary No 369 (80) 383 (93) <0.001 140 (56) 256 (56) 0.967 30 (75) 23 (77) 0.872 Yes 92 (20) 28 (7) 109 (44) 198 (44) 10 (25) 7 (23) Missing 1 (0) 4 (1) 161 (39) 339 (43) — — (y)pT-stage 0 6 (1) 15 (4) 0.004 0 (0) 7 (1) <0.001 0 (0) 0 (0) 0.823 1 11 (2) 4 (1) 21 (5) 10 (1) 1 (3) 0 (0) 2 74 (16) 47 (12) 46 (12) 65 (9) 4 (10) 4 (13) 3 327 (72) 259 (69) 263 (67) 462 (66) 30 (77) 23 (77) 4 38 (8) 51 (14) 60 (15) 153 (22) 4 (10) 3 (10) Missing 6 (1) 39 (9) 20 (5) 96 (12) 1 (2) 0 (0) (y)pN-stage 0 196 (43) 139 (37) 0.202 185 (46) 246 (32) <0.001 17 (44) 13 (43) 0.893 1 170 (38) 151 (40) 155 (38) 307 (40) 17 (44) 12 (40) 2 87 (19) 83 (22) 64 (16) 213 (28) 5 (13) 5 (17) Missing 9 (2) 42 (10) 6 (1) 27 (3) 1 (2) 0 (0) Differentiation grade pCR 5 (2) 13 (4) 0.189 — — 0 (0) 0 (0) 0.733 G1 6 (2) 6 (2) — — 9 (23) 7 (23) G2 267 (93) 266 (89) — — 28 (72) 20 (67) G3 9 (3) 15 (5) — — 2 (5) 3 (10) Missing 175 (38) 115 (28) 410 (100) 793 (100) 1 (2) 0 (0) Disease-free interval in months (median [IQR]) 11.0 [0.0–22.8] 0.0 [0.0–0.5] <0.001 12.0 [0.0–24.8] 0.0 [0.0–5.0] <0.001 5.8 [0.0–14.5] 0.0 [0.0–14.8] 0.334 Number of CRLM (median [IQR]) 1.0 [1.0–2.0] 3.0 [1.0–5.0] <0.001 1.0 [1.0–2.0] 2.0 [1.0–4.0] <0.001 1.0 [1.0–2.0] 2.0 [1.0–3.0] 0.096 Diameter of largest CRLM in cm (median [IQR]) 3.2 [2.1–4.8] 2.2 [1.3–3.7] <0.001 3.0 [2.1–5.0] 2.5 [1.6–4.2] <0.001 3.0 [2.5–4.6] 2.5 [2.0–3.6] 0.100 Preoperative CEA in μg/l (median [IQR]) 11.0 [4.1–29.0] 19.0 [5.3–74.0] <0.001 9.8 [3.8–31.2] 9.5 [3.9–35.3] 0.577 7.4 [2.3–23.9] 17.2 [4.8–58.5] 0.078 Clinical risk score Low risk (0–2) 331 (75) 177 (48) <0.001 286 (74) 346 (47) <0.001 27 (69) 15 (52) 0.142 High risk (3–5) 111 (25) 190 (52) 100 (26) 393 (53) 12 (31) 14 (48) Missing 20 (4) 48 (12) 24 (6) 54 (7) 1 (2) 1 (3) Extrahepatic disease No 429 (93) 350 (84) <0.001 362 (88) 651 (82) 0.005 38 (95) 27 (90) 0.421 Yes 33 (7) 65 (16) 48 (12) 142 (18) 2 (5) 3 (10) Resection margin status R0 410 (90) 329 (80) <0.001 368 (91) 675 (86) 0.009 37 (92) 29 (97) 0.457 R1 48 (10) 84 (20) 35 (9) 109 (14)