Lymphocyte Count

In order to represent a weight of the interaction between inflammatory pro-tumour populations (i.e. neutrophils, platelets and monocytes) and anti-cancer immune populations (i.e. lymphocytes), PIV was calculated as: [neutrophil count (10³/mmc) × platelet count (10³/mmc) × monocyte count (10³/mmc)]/lymphocyte count (10³/mmc). Maximally selected rank statistics method for PFS was used to find an optimal cut-off value^{13 (link)} to stratify patients in low PIV vs high PIV. NLR was calculated as: neutrophil count (10³/mmc)/lymphocyte count (10³/mmc). NLR was defined high if >3, platelet count was defined high if >310 × 10³/mmc and monocyte count was defined high if >0.5 × 10³/mmc based on literature data.^{6 (link)–8 (link)} SII was calculated as [neutrophil count (10³/mmc) × platelet count(10³/mmc)/lymphocyte count (10³/mmc) and defined high if >730 based on literature data.^{10 (link)}Fisher exact test, Chi-square test, Mann–Whitney U test or Kruskal-Wallis test, as appropriate, were used to analyse the association between baseline PIV and the other clinicopathological characteristics. PFS was defined as the time from randomisation to disease progression or death from any cause. OS was defined as the time from randomisation to death from any cause. Generalised boosted regression was used to screen the association of PIV and the other IIBs with PFS and OS.^{14 (link),15 (link)} Further survival analyses were performed using the Kaplan–Meier method and the Cox proportional hazards regression models. All the variables showing a P below the significance threshold in the univariate models were included in a multivariable model. The variables showing a P below the significance threshold in the multivariable models were considered to be independent prognostic factors. All tests were 2-sided with a significance threshold of 0.05. Statistical analyses were performed using the R (version 3.5.0) and R Studio (version 1.1.447).

We performed co-localization analysis on QTLs in the eQTL Catalogue against GWAS summary statistics from 14 studies downloaded from the IEU OpenGWAS database in VCF format^{98 ,99} . Our analysis included summary statistics for inflammatory bowel disease (IBD) and its two subtypes (Crohn’s disease (CD) and ulcerative colitis (UC))^{100 (link)}; rheumatoid arthritis (RA)^{101 (link)}, systemic lupus erythematosus (SLE)^{102 (link)}, type 2 diabetes (T2D)^{103 (link)}, coronary artery disease (CAD)^{104 (link)}, LDL-cholesterol¹⁰⁵ , four blood cell type traits (lymphocyte count (LC), monocyte count (MC), platelet count (PLT), mean platelet volume (MPV))^{34 (link)} and two anthropometric traits (height, body mass index (BMI)) from the UK Biobank¹⁰⁵ . The variant coordinates of the GWAS summary statistics were lifted to the GRCh38 reference genome using CrossMap^{57 (link)}. We used v.3.1 of the coloc R package^{106 (link)}. All analysis steps are implemented in the v.21.01.1 of the eQTL Catalogue/co-localization workflow (see URLs).
We used our uniformly processed GTEx summary statistics together with all the other summary statistics from eQTL Catalogue release 3.1. For all eQTL and GWAS dataset pairs, we performed co-localization in a ±200,000 window around each of the 62,837 fine-mapped eQTL credible set lead variants (see Statistical fine mapping above). This ensured that co-localization was also performed separately for multiple independent eQTLs of the same gene and co-localization results were obtained in datasets in which no significant eQTL was detected for a particular gene. However, as we did not use masking or conditional analysis, many secondary eQTL co-localizations could still have been missed^{18 (link),107 (link)}. Inspired by the study by Barbeira et al.³ , we summarized strong co-localizations (PP4 ≥ 0.8) at the level of approximately independent LD blocks^{35 (link)}. Positions of approximately independent LD blocks were obtained from Berisa and Pickrell^{35 (link)} and converted to GRCh38 coordinates using CrossMap^{57 (link)}. If the co-localization cis window overlapped two or more LD blocks, then the co-localizing QTL was assigned to the LD block where the QTL lead variant was located. We defined an LD block to harbor a novel co-localization signal if there was no co-localization detected within that LD block in any of the GTEx tissues. We further excluded datasets with small sample sizes (n < 150) due to their low power to detect co-localizations.
As transcript usage, exon expression and txrevise contained many more redundant phenotypes (for example, multiple exons of the same gene), we limited co-localization analysis for those molecular traits to the significant lead QTL variants in each dataset only (false discovery rate (FDR) < 0.01), using the same ±200,000 cis window as above. To make the co-localization signals comparable across quantification methods, we also performed co-localization analysis for gene expression using significant lead QTL variants as we did for the other three quantification methods. We only included QTL and complex trait pairs with strong evidence of co-localizations (PP4 ≥ 0.8) in our analysis and summarized the results at the level of independent LD blocks, as described above. The number of LD blocks for which we detected at least one co-localizing QTL with each quantification method was visualized using the upsetR v.1.4.0 R package^{108 (link)}.

Data, including age, sex, histological subtype, stage, smoking status, chemotherapy regime, Eastern Cooperative Oncology Group Performance Status (ECOG PS) scores, routine blood parameters and biochemical profiles, were collected retrospectively from individual medical case notes, electronic patient records and pathology reports. Blood samples were obtained and assayed within 2 weeks before chemotherapy. CONUT scores were summarized using the serum albumin concentration, peripheral lymphocyte counts and the total cholesterol concentration, as described in Table 1. The following formula was used to calculate PNI and SII. PNI: albumin (g/L) × total lymphocyte count × 10⁹/L. SII: platelet count × neutrophil count/lymphocyte count [9 (link), 16 (link)]. Follow-up was performed every 3 months. All patients were monitored either until July 2020 or until death. The median follow-up time was 24 months (range, 3–75 months). Progression-free survival (PFS) was defined as the interval from treatment initiation until disease progression or death. Overall survival (OS) was defined as the interval from treatment initiation until the date of death or the date of last follow-up for patients who had not died before the censor date.Table 1

The CONUT scoring system

Parameters	Degree of undernutrition
	Normal	Light	Moderate	Severe
Serum albumin (g/dL)	≥ 3.5	3.0–3.4	2.5–2.9	< 2.5
score	0	2	4	6
Total lymphocyte count (mm3)	≥ 1600	1200–1599	800–1199	< 800
score	0	1	2	3
Total cholesterol  (mg/dl)	≥ 180	140–179	100–139	< 100
score	0	1	2	3
CONUT score (total)	0–1	2–4	5–8	9–12

CONUT Controlling nutritional status

The Mann-Whitney U-test and Fisher's exact test were used for age- and sex- matching, respectively. The Mann-Whitney U-test was used to evaluate significance in the differences between antibody levels in different groups. Simple linear regression and point-biserial correlation were used to analyze the correlation between antibody levels and lymphocyte count. The GraphPad Prism v8·00 for Windows software program was used to perform the statistical analyses (GraphPad Software, La Jolla, California, USA).