Methods

Patient enrollment

Patients who met the following inclusion criteria were eligible for this study: (1)

they had pathologically confirmed de novo metastatic BRAF V600E mutated CRC,

which referred to distant metastases at diagnosis; (2) they had colectomy for CRC; and

(3) they had MSS CRC. Patients who received systemic chemotherapy or targeted

therapy before colectomy were ineligible for this study. We enrolled patients from two

medical centers: NTUH and Kaohsiung Chang Gung Memorial Hospital (KCGMH,

Kaohsiung, Taiwan). At NTUH, we searched the medical records on the BRAF mutation

examination from 2010 to 2017. The codes for BRAF mutation analyses included

000W0406 for expanded RAS+BRAF mutation, 000W0389 for KRAS and BRAF

mutations, and 000W0404 for BRAF mutation alone. Then we could identify patients

with de novo metastatic BRAF V600E mutated CRC. In addition, we included the

patients who met the inclusion/exclusion criteria from our previous cohort (2005-2013).

Finally, another cohort from KCGMH was also enrolled. We collected patients’

clinicopathological characteristics, in terms of age, sex, histology, tumor grade, disease

stage at diagnosis, primary tumor location, systemic chemotherapy, targeted therapy,

and survival time from electronic medical records. Patients were recognized to have

right-sided tumors if the primary tumors were located from the cecum to the splenic

flexure of the transverse colon. Patients had left-sided tumors if the primary tumors

were located from the descending colon to the rectum. The definition of first-line triplet

chemotherapy was the anti-cancer therapy consisting of oxaliplatin, irinotecan, and one

of the fluoropyrimidines. The first-line doublet referred to the anti-cancer therapy

consisted of one of the fluoropyrimidines and either oxaliplatin or irinotecan. This study

was approved by the Institute Review Board of NTUH and KCGMH.

BRAF V600E mutational analysis

The methods for detecting BRAF V600E mutation at NTUH include PCR and

direct sequencing (ABI 3730 sequencer (Applied Biosystems, Foster City, CA, USA).

The primer pairs (5’-TCA TAATGCTTGCTCTGATAGGA-3’ and 5’ - GGCCA

AAAATTTAATCAGTGGA-3’) covering BRAF exon 15 were used. On the other hand,

BRAF V600E mutation was identified by Roche BRAF/NRAS Mutation Test (Roche

Diagnostics, IN, USA) at KCGMH following the manufacturer’s instructions. We

performed BRAF mutation subtype classification (BM1 and BM2) by linear regression

classifier as described in a previous study⁷⁹.

MSI analysis

The method for MSI testing and the definitions of MSI-high and MSS have been

described in chapter 2.

Transcriptomic analysis

We extracted total RNA from the FFPE samples of primary colon tumors derived

from colectomy using Qiagen’s miRNeasy FFPE Kit (Qiagen, Valencia, CA, USA).

Macrodissection of tumor slides was performed before RNA extraction. Then we used

NanoDrop 2000 at 220–320 nm (Thermo Scientific) to measure 260/280 ratios to

determine RNA quality and quantity. Bioanalyzer (model 2100, Agilent Technologies,

Santa Clara, CA, USA) was used to measure nucleic acid fragmentation. Finally, the

gene expression was evaluated by nCounter PanCancer Immune Profiling Panel

(NanoString, Seattle, WA, USA). This panel consists of 770 genes and is widely used to

investigate TME. We processed raw data as previously described⁹⁰. Gene counts for

each sample were normalized first by trimmed mean of M-values method⁹¹ and then

converted to log2-transformed counts per million later for clinical analysis.

Development of complement scores

We calculated pairwise expression correlation in complement genes according to

the similarity metric of Pearson’s correlation coefficient⁹². The results were regarded as

an input for hierarchical clustering. Then the complement genes were classified into

three co-expression modules, and the clustering results were shown by a heatmap

symmetrical along the diagonal line. The complement score of each sample was

determined as the average gene expression in module 2. A high complement score was

defined when the complement score was higher than the median score; in contrast, a

low complement score was defined when the complement score was lower than the

median score.

Evaluation of immune cell abundance

The immune cell abundance in the TME of each tumor was measured according to

its expression profiles. The simple average of marker genes’ expression values of each

immune cell was calculated as cell-type scores. The reference marker genes were

determined by Danaher et al.⁹³. We also used the CIBERSORT algorithm with the

LM22 signature matrix to measure cell abundance⁹⁴. The phenotypes of macrophages

(M1 and M2) were defined based on the mean expression values of gene signatures⁹⁵.

The genes correlated to M1 macrophage consist of CD40, CD86, CCL5, CCR7, CXCL9,

CXCL10, CXCL11, IL1A, IL1B, IL6, IRF1, IRF5, IDO1, and KYNU, and the M2

macrophage-associated genes are CD276, CCL4, CCL13, CCL18, CCL20, CCL22,

CTSA, CTSB, CTSC, CTSD, CLEC7A, FN1, IL4R, IRF4, LYVE1, TGFB1, TGFB2,

TGFB3, TNFSF8, TNFSF12, MMP9, MMP14, MMP19, MSR1, VEGFA, VEGFB, and

VEGFC.

External cohorts

The first external cohort was colon and rectal adenocarcinomas in the TCGA

dataset⁹⁶. Genomic and transcriptomic data of these tumors were obtained from

Genomic Data Commons through the R package of TCGA bio-links, and these tumors’

microsatellite status was obtained according to the method described preciously⁹⁷. The

second external cohort was GSE39582 dataset⁹⁸. Complement scores of all these CRC

tumors from two external cohorts were calculated, and tumors were determined to have

high or low complement scores based on the abovementioned definitions.

Immunohistochemical staining

The collected FFPE tumor tissues were cut into 5-µm sections, and then these

sections were deparaffinized by EZ prep (Ventana Medical Systems, Inc., Tucson, AZ,

USA). These sections were also pretreated with Cell Condition 1 solution for 32

minutes (Ventana Medical Systems, Inc.). We incubated the slides with anti-human C4d

(clone A24-T, DB Biotech) and used the automated Ventana Benchmark XT for 1:200

titration (Ventana Medical Systems, Inc.). Labeling was measured using the Optiview

DAB Detection Kit (Ventana Medical Systems, Inc.). All sections were counterstained

with hematoxylin in Ventana reagent, and the positive controls were the human FFPE

tonsil tissues.

C4d is generated after C4 splitting during the activation of the complement

pathway. C4d is a surface-bound protein that binds to tissue near the activation site

because its covalent bond does not break spontaneously⁹⁹. Thus, C4d is recognized as a

marker of complement activation. We categorized CRC into three groups based on the

percentages of C4d deposits on the tumor cell membrane. C4d-IHC 0 (weak) referred to

<1% of neoplastic cells, C4d-IHC 1 (intermediate) was defined as 1%–30% of

non-neoplastic cells, and C4d-IHC 2 (high) was defined as >30% of non-non-neoplastic cells.

Semi-quantification was evaluated by two independent pathologists.

Statistical analysis

Patients’ clinical and pathological data were summarized as percentages. Survival

analyses among groups, including PFS, OS, and progression-free interval (PFI), were

evaluated using the Kaplan-Meier method. A p-value of less than 0.05 in the log-rank

test was considered significant. We developed a Cox proportional hazards regression

model for survival analysis with the coxph function implemented in the R package. PFS

in this study was defined as the period from the date of diagnosis to the date of

radiographically confirmed progression of disease or death during first-line therapy. OS

in this study was defined as the time from diagnosis to death. PFI was specifically used

in TCGA and GSE39582 datasets to evaluate clinical outcomes^96,98. PFI was defined as

the time from the date of diagnosis to the date of the first evidence of disease

progression, locoregional recurrence, new primary tumor, distant metastasis, or death.

Only variables with p < 0.1 were included in the multivariate analysis. We used

Benjamini and Hochberg multiple-testing correction method to adjust multiple tests.

The statistical analyses and visualization in this study were performed in an R

environment (version 3.6.1).