Title: Anti-tumour necrosis factor-alpha response associated with combined CD226 and HLA-

DRB1*0404 haplotype in rheumatoid arthritis.

Authors: Gibson DS1*†, McGeough CM1, 5†, Watterson S1, Blayney J6, Wright GD2, Pendleton A2,

Gardiner P3, Small D3, Eakin AJ1, Ahmed T1, Murray HA4, Latten MJ4, Crockard MA4, Lamont JV4,

Zhang SD1, Bjourson AJ1.

Institutional Address:

1. Northern Ireland Centre for Stratified Medicine, Biomedical Sciences Research Institute, University

of Ulster, Altnagelvin Hospital, Londonderry, UK.

2. Department of Rheumatology, Musgrave Park Hospital, Belfast, UK.

3. Department of Rheumatology, Altnagelvin Hospital, Londonderry, UK.

4. Randox Laboratories Ltd., Crumlin, UK.

5. Almac Diagnostics Ltd., Craigavon, UK.

6. Institute for Health Sciences, Queens University Belfast, Belfast, UK.

*corresponding author d.gibson@ulster.ac.uk

†DSG and CMcG contributed equally to this work.

Word Count: 2997 excluding abstract, title, authors, references, legends.

Page 1 of 25

Abstract (231 words)

Objective

Predicting response to anti-tumour necrosis factor alpha (anti-TNFα) drugs at baseline remains an

elusive goal in rheumatoid arthritis (RA) management. The purpose of this study was to determine if

baseline genetic variants of PTPRC, AFF3, myD228, CHUK, MTHFR1, MTHFR2, CD226 and a

number of KIR and HLA alleles could predict response to anti-TNFα in rheumatoid arthritis patients.

Methods

Peripheral blood samples were collected from 238 RA patients treated with anti-TNFα drugs.

Genotyping was performed using biochip array technology by Randox Laboratories Ltd. and

sequence specific polymerase chain reaction. Linear regression analysis was performed to investigate

the role of these genotypes in predicting response to treatment, as defined by European League

Against Rheumatism (EULAR) response classification and absolute change in disease activity score

(DAS28).

Results

Of 238 RA patients analysed, 50.4% received adalimumab, 29.7% received etanercept, 14.8%

received infliximab, 3.4% certoluzimab and 1.7% golimumab. The MTHFR1 variant rs1801133 was

significantly associated with the EULAR response, p=0.044. Patients with the HLA-DRB1*0404 allele

displayed a significantly larger reduction in DAS28 compared to non-carriers (mean -2.22, -1.67

respectively, p=0.033). CD226 rs763361 was the only SNP variant significantly associated with

∆DAS28 (p=0.029)

Conclusion

This study has investigated individual allele associations with reductions in DAS28 across a range

of anti-TNFα treatments. A combined predictive model indicates that patients with the HLA-

DRB1*0404 allele and without the CD226 rs763361 polymorphism exhibit the largest reduction in

DAS28 after anti-TNFα treatment.

Page 2 of 25

Keywords: anti-TNFα, rheumatoid arthritis, response prediction, genotype, polymorphism

Background

Anti-tumour necrosis factor alpha (anti-TNFα) blocking agents are effective at reducing disease

activity measures in approximately 70% of rheumatoid arthritis (RA) patients (1). As TNF is a key

driver of joint inflammation, in those who respond, anti-TNFα biologic drugs reduce immune cell

infiltration into the joint and diminish joint destruction. Now that there are several classes of biologic

drug available for severe RA, there is a pressing need to find a widely accepted test that will help the

clinician predict response and select the most appropriate therapy for each patient (2). A number of

studies have assessed potential clinical and biological predictors of response, but in many cases

biomarkers that appeared to be promising were not confirmed as clinically relevant markers in further

studies.

A lower baseline health assessment questionnaire score and the concurrent use of methotrexate

are associated with a good response to anti-TNFα treatment (1). A high body mass index, low

baseline disease activity and smoker status have all been associated with reduced rates of response

to anti-TNFα drugs (3-5).The presence of rheumatoid factor and/or anti-citrullinated protein antibodies

(ACPA) has been associated with response to TNF-inhibitors in some but not all studies (1, 5-7). A

pilot study in 170 RA patients demonstrated how clinical factors could be combined with a serum

biomarker (serum myeloid-related protein, MRP8/14) in a treatment algorithm capable of predicting

anti-TNFα response (8).

Many single nucleotide polymorphism studies have demonstrated associations with anti-TNFα

treatment response. Genetic variants of components which map to T cell function associate with

response, including IRAK 3, which negatively regulates toll-like receptor (TLR) signalling, conserved

helix-loop-helix ubiquitous kinase (CHUK) and myeloid differentiation primary response protein

(MyD88), which activate or inhibit nuclear factor-κB (NF-kB) signalling (9, 10). MyD88 and CHUK

were previously associated with etanercept response in RA patients (7). A contradictory study found

no response association with genetic variants of CD226, AF3/FMR2 family, member 3 (AFF3) in

Page 3 of 25

addition to CHUK and MyD88 (11). Genetic variants of the mitogen-activated protein kinase (MAPK)

signalling pathway components also showed associations with infliximab and adalimumab responders

in RA (12).

The shared epitope susceptibility loci have also formed the focus of anti-TNFα response association

studies. Studies of UK and Swedish RA patients indicate no clear association with response, though

these studies generalise the various human leukocyte antigen (HLA) DRB1 haplotypes rather than

studying allele specific associations (13-15). A study which focused on dissecting shared epitope

phenotypes established that valine at position 11 of HLA-DRB1 is associated with improved response

to anti-TNFα therapy in RA (16).

In a previous pilot study of RA by our group, the presence of specific haplotypes of genes central to

T cell function Killer cell immunoglobulin-like receptor 2DS2 (KIR2DS2) and HLA-C group 1/2

(homozygous), was observed as being associated with anti-TNFα response (17). We therefore

designed an expanded study to independently validate whether specific KIR and HLA haplotypes and

several promising genetic variants could predict response to anti-TNFα treatment in a cohort of

biologic naive UK RA patients.

Methods

Patient selection

The following inclusion criteria were used for the current study: 1). RA patients fulfilling ACR/EULAR

diagnostic criteria (18), 2). Treated with anti-TNFα treatment as part of routine clinical practice, 3).

Fulfilled the BSR 2001 criteria for anti-TNFα therapy, 4). Had a DAS28 score of >5.1 when originally

assessed for treatment (before baseline), 5). Reached 6 months of follow-up. Patients who stopped

anti-TNFα temporarily during first six months or discontinued therapy prior to the 6 month follow up for

reasons other than inefficacy were excluded. Two hundred and thirty eight eligible RA patients were

recruited from rheumatology biologic clinics at Altnagelvin Hospital, Londonderry and Musgrave Park

Hospital, Belfast, Northern Ireland. Office for Research Ethics Committees Northern Ireland (ORECNI)

approval (11/NI/0188) was obtained for the study. Informed consent was obtained for all participants

in the study, allowing for publication of anonymised clinical data.

Page 4 of 25

Patient recruitment, sample collection, clinical information

The study was supported by a patient advisory group who met regularly throughout the study to

advise on study design, recruitment literature and results dissemination. Eligible patients were invited

by mailed patient information sheets, a minimum of 48 hours before a routine care appointment.

Additional blood samples were obtained from consenting patients who were either about to

commence, or had been on an anti-TNFα treatment in the past. Blood samples were processed by a

silica based extraction kit to isolate genomic DNA (DNAEasy, Qiagen Inc.). DNA samples were

aliquoted and stored at -80°C until analysis. Clinical and demographic information was extracted from

medical records and clinic databases after written informed consent. Rheumatoid factor (RF; <30

IU/mL = negative, 30-50 IU/mL = weak positive, >50 IU/mL = positive) and anti-cyclic citrullinated

peptide (anti-CCP; >10U/ml = positive, 7-10U/ml = equivocal, <7U/ml = negative) tests were

performed by turbidimetry and automated enzyme immunoassay, respectively. Disease activity was

compiled for baseline and 6 months of treatment with anti-TNFα, using DAS28-ESR (19). Following 6

months of treatment, the patients were assigned a moderate responder, good responder or non-

responder status, according to EULAR criteria (20). The main demographic and clinical features of the

patients are shown in Table 1.

Genotyping

All genotyping was performed by custom RheumastratTM biochip array technology (Evidence

AnalyserTM, Randox Laboratories Ltd.). Genotyping was confirmed by PCR-SSOP method (17).

Positive controls of known KIR, HLA or single nucleotide polymorphism (SNP) genotype, were

included in the typing procedure. DNA was typed for the presence or absence of previous response

associated framework KIR genes: KIR2DS2 (activator) and KIR2DL2 (inhibitor). HLA-DRB1 typing

was performed on the following shared epitope alleles: *03, *0101, *1001, *0401, *0104 and *0404. A

modified version of the HLA-C typing method was used to define the HLA-C1 and C2 groups using

probe C293 and C291, respectively (17).

Single nucleotide polymorphisms previously published as associated with therapeutic response and

disease severity were typed for the following gene loci (in brackets): HLA-DR/BTNL2 (rs1980493),

Page 5 of 25

protein tyrosine phosphatase, receptor type C (PTPRC) (rs10919563), AFF3 (rs10865035), CD226

(rs763361), myD88 (rs7744), CHUK (rs11591741), methylenetetrahydrofolate reductase 1 (MTHFR1)

(rs1801133), MTHFR2 (rs1801131) (7, 21-25).

Statistical methods and analysis

The study size was based on power calculations that indicated that an overall sample size of 150

patients across 3 groups would provide > 80% power (at the conventional significance level

alpha=0.05) to identify a genetic factor with a small effect size (Cohen’s f-squared = 0.10).

The significance of the differences in proportions of responders and non-responders exhibiting a

specific genotype was assessed using Fisher’s exact test. For the numeric measures including

DAS28 scores at baseline, DAS28 score changes at month 6 (∆DAS28), Health Assessment

Questionnaire scores at baseline, t-based statistics were used to assess the difference between two

means, one-way ANOVA for multiple means, and Pearson correlation for assessing correlations

between numerical variables. Whereas the normality assumptions underpinning those parametric

tests above do not firmly hold, non-parametric counterparts of those techniques (eg. Kruskal-Wallis

test instead of one-way ANOVA) were applied (as indicated in Table 1) . In particular for the DAS28

scores at baseline, at month 6 and their difference (∆DAS28), the normality assumptions were well

supported by the data, hence validating the parametric tests used and subsequent linear regression

analyses. For the other numerical measures in Table 1, Age, baseline TJC, SJC, CRP, and HAQ, the

non-parametric Kruskal-Wallis tests were used, and median and interquartile range (IQR) instead of

mean and standard deviation are reported. Using the software R, A series of systematic linear

regression analyses were used to construct a most appropriate models consisting of significant

predictors (detailed in the Results section). All tests were two-sided unless otherwise stated. Where

applicable, Adjustments for multiple testing were made using Holm’s method.

Results

Patient demographics

Page 6 of 25

There was no significant difference among the three response groups (ie, good, moderate, and non-

responder) of patients with respect to the distribution of age (p = 0.74), gender (p = 0.32) or

concomitant conventional DMARD use (p=0.70) (Table 1 study cohort characteristics). Baseline

tender and swollen joint counts were significantly lower in non-responders, whereas baseline CRP

was significantly elevated in moderate responders; both differences were statistically significant

(p=0.027, p=8.5E-04, p=0.063 respectively). The mean change in DAS28 differed significantly

between each EULAR response group, with non-responders at 0.4 ± 1.0, moderate responders -1.8 ±

0.8 and good responders -3.1 ± 1.0 (p=4.10E-45). For patients who had laboratory data available, 149

or 73.8% of those tested were rheumatoid factor positive at the start of the study and 94 or 72.9% of

those tested were anti-cyclic citrullinated peptide (anti-CCP) antibody positive.

238 patients received a combination of conventional DMARD (cDMARD) and anti-TNFα drugs. Of

the five anti-TNFα drugs prescribed in the study population, adalimumab was prescribed for 119

(50.4%), etanercept for 70 (29.7%), infliximab for 35 (14.8%), certoluzimab for 8 (3.4%) and

golimumab for 4 (1.69%) (Supplementary Table 1 of anti-TNFα drugs prescribed across study cohort).

There was no significant difference in treatment outcome between the five anti-TNFα drugs when

comparing patients who did or did not receive concurrent cDMARDs such as methotrexate (ANOVA, p

= 0.093).

Baseline characteristics (gender, HAQ and genotype vs baseline DAS28)

Baseline DAS28 was found to be significantly higher in patients with HLA-DRB1*0404 haplotype

(p=0.038; absent mean DAS28 5.30 ±1.235; present mean DAS28 5.75±1.151; Figure 1A,

Supplementary Table 2). Gender and baseline cDMARD did not exert a statistically significant effect

on change in DAS28 over 6 months of treatment (∆DAS28) in the study RA population (Table 2).

Since strength of association of a particular genotype with ∆DAS28 could be influenced by gender,

baseline DAS28 and concurrent cDMARD use, adjustments considering these factors were necessary

in later predictive model tests. As has been observed in previous studies, baseline DAS28 was

significantly correlated with baseline HAQ (p0.005; R=0.227; n=154; Figure 1B).

Individual factor association with response to anti-TNFα treatment

Page 7 of 25

Using the EULAR classification system none of the shared epitope alleles were strongly associated

with response to anti-TNFα (HLA-DRB1*0404 p=0.059, Supplementary Table 3). The MTHFR1 SNP

was weakly associated with the EULAR response (p=0.044), but the CD226 SNP was not (p=0.202).

The associations between presence or absence of individual alleles on ∆DAS28 were investigated

(Supplementary Table 4; Figure 2). Notably, presence of the HLA-DRB1*0404 allele compared to

absence was significantly associated with a larger reduction in DAS28 (mean -2.22 and -1.67

respectively, p=0.033). CD226 SNP was the only gene variant significantly associated with ∆DAS28

(p=0.029) (Supplementary Table 4). The mean reduction in DAS28 was greatest for the CD226

homozygous wild type TT variant (-2.35), compared to heterozygous carrier genotype TC (-1.57) and

homozygous genotype CC (-1.85).

Multi-factor regression models

As previous reports indicate that ∆DAS28 could be influenced by gender, baseline DAS28 and

concurrent cDMARD use, and HAQ, we thus investigated whether these four factors are significantly

associated with ∆DAS28 in our cohort of patients (1). We fit a linear regression model using ∆DAS28

as dependent variable, and the four factors mentioned here as independent variables. A complicated

full model with all possible pair-interaction terms was explored. However, with this 11-term model

(one intercept term, 4 terms for individual variables, and 6 interaction terms), only the baseline term

(recDAST0) would be significant (data not shown). We therefore removed the interaction terms with

higher p-values, and started with only the interaction term between the two categorical variables

(Gender and concurrent cDMARD). However the gender-cDMARD interaction term was still found to

be non-significant (p-value = 0.26), subsequently a simplified regression model without interaction

terms was fitted. The results of this regression analysis are shown in Supplementary Table 5. As can

be seen from table, gender (p = 0.99) and HAQ (p=0.52) were not significantly associated with

∆DAS28. Both factors were excluded from all subsequent analysis.

As expected, baseline DAS28 score was a highly significant predictor of response (p < 0.001), while

concurrent cDMARD therapy was only of borderline significance (p =0.051). Subsequently, baseline

DAS28 was built into a base model, with cDMARD and all the genetic factors added individually to the

Page 8 of 25

base model to examine their association with ∆DAS28. Briefly, a series of linear regression models

were fitted with baseline DAS28, plus one genetic factor or one other factor. The other factors

considered here included RF status, anti-CCP status, the type of anti-TNFα used and concurrent

cDMARD use. Screening through all the genetic factors and the other listed factors, only CD226 was

found to contribute significantly (at the level of alpha=0.05) to ∆DAS28 after adjusting the effect of

baseline DAS28 score (Table 3; Figure 3B). The HLA-DRB1*0404 haplotype, although found to be

significantly associated with ∆DAS28 in the individual factor analysis above, was no longer significant

after correcting for the effects of baseline DAS28 score.

In summary, the baseline DAS28 score and the CD226 status can be combined in a model to

predict anti-TNFα response in this cohort of RA patients. The results of this model are shown in Table

5, and the effects of baseline DAS28 and CD226 genotypes are depicted in Figure 3. As can be seen

from this figure, the homologous CD226 genotype 11 conforms to the best response, while patients

with either genotype 12 or 22 have significantly worse response than genotype 11.

Given the finding about HLA-DRB1*0404 in the individual factor analysis, we further investigated

whether it might still have predictive value in the base model with baseline DAS28 and CD226 as

built-in predictors. First, HLA-DRB1*0404 was added to this base model without interaction with

CD226. In this simple interaction-free model, HLA-DRB1*0404’s effect on ∆DAS28 is just short of

statistical significance (p =0.051). Secondly, we considered the interaction between CD226 and HLA-

DRB1*0404 in the model, and found interesting results. There are significant differences among the 6

genotypes of CD226-HLA-DRB1*0404 combinations; Homozygous absence of the CD226 rs763361

SNP and presence of HLA-DRB1*0404 allele represents the most responsive genotype, which is

significantly better than most of other 5 combinations (Figure 3C).

Discussion

Rather than grouping all HLA-DRB1 genotypes together as in previous studies (13), this study has

for the first time tested individual allele associations with ∆DAS28 across a range of anti-TNFα

treatments. This is also the first study to report a combined predictive model which indicates that

Page 9 of 25

patients with presence of HLA-DRB1*0404 and absence of CD226 SNP rs763361 exhibit the largest

reductions in DAS28 after anti-TNFα treatment.

HLA-DRB1*0404 carriers are known to be predisposed to a more severe arthritis phenotype and a

higher disease activity (26, 27). In the current study patients with the HLA-DRB1*0404 haplotype did

manifest significantly elevated baseline disease activity. Potter et al. in a replication study were

unable to verify the association of shared epitope genotypes with ∆DAS28 in a study of anti-TNFα

treatment (13). It is therefore of particular interest that the current study, confirmed that the presence

HLA-DRB1*0404 was independently associated with a significantly larger drop in disease activity after

anti-TNFα treatment. The association of the *0404 and *0101 alleles with response has been

observed previously with etanercept, though intriguingly only when combined with lymphotoxin-α-TNF

polymorphisms (28). The influence of TNFα and TNF receptor gene polymorphisms upon response to

anti-TNFα response have been extensively studied. The TNF SNPs 489G/A (rs80267959), -308G/A

(rs1800629) and -857C/T (rs1799724) are particularly associated with response to etanercept (28-30).

Consistent with previous work, a significant association between CHUK (rs11591741) SNP with

etanercept response was observed in the subgroup of 70 RA patients who received this anti-TNFα,

though there was no significant association MyD88 SNP (rs7744) (7).

The presence of CD226 SNP rs763361 in the study population was associated with significantly

reduced responses to anti-TNFα treatment. A similar association between rs763361 and anti-TNFα

response in RA has already been reported by Tan et al., though as they noted it had only modest

strength in a predictive model alone (23). We have therefore attempted in a combined predictive

model to control for the possibility that larger changes in ∆DAS28 are not solely due to higher baseline

DAS28 in HLA-DRB1*0404 carriers. In the predictive model, only the CD226 SNP contributes to

significant drops in disease activity post ant-TNF, once the potentially confounding effects of baseline

DAS28 are corrected.

Biological significance

It is thought that anti-TNFα targeting of inflammatory cells with membrane bound TNF enhances

antibody dependent cellular cytotoxicity (ADCC) by macrophages and natural killer cells (31). Cell

Page 10 of 25

surface activating and inhibitory killer cell immunoglobulin-like receptors regulate natural killer cell

functions via HLA class I molecule interaction (32). So it is reasonable to postulate that although the

*0404 allele confers higher disease risk and activity, it may also positively modify ADCC mediated

apoptosis and clearance by natural killer cells.

CD226 is involved in the effector functions of T helper cells and peripheral T cells exhibit increased

CD226 expression in rheumatoid arthritis (33). The rs763361 SNP located in exon 7 of CD226 confers

a Gly307Ser change within the cytoplasmic tail of the CD226 receptor. This variant is strongly

associated with susceptibility to multiple autoimmune conditions including type 1 diabetes, multiple

sclerosis and RA (34). The biological consequence of the variant remains unclear, though it has been

hypothesized that downstream effects on phosphorylation at Ser329 may be affected, which is

required for cell activation via LFA1 (35).

Clinical significance

The ability to correctly predict responders for high cost biologic treatments remains a lofty goal.

Previously mentioned studies have reported a number of promising genotypes, but many observe that

though associations may be strong or statistically significant, they may be of limited clinical benefit in

managing patients. The ability of our model to correctly predict true responders (test sensitivity) is

poor (23% HLA0404 and 19% CD226; Supplementary Table 6). However, the ability to distinguish

future responders with a positive HLA0404 test and a negative CD226 test was good with positive

predictive values of 82% and 83%, respectively. This ‘future responder’ HLA0404-CD226 combined

genotype represents 17% of the study population.

Other observations

Although the sample size could have been larger, over 200 RA patients were studied and therefore

our study had adequate statistical power. It is also interesting to note that responders (moderate and

good) appeared to have higher baseline tender joint counts (TJC) compared to non-responders. It

may suggest that the there is a different disease process associated with those that are responding

versus those not responding. Although the *0404 allele does not appear to have a significant

Page 11 of 25

influence upon baseline TJC, further investigation of the biological pathways associated with TJC

factors may be worthy of further study.

Conclusions

Further association studies of combined haplotypes and other ‘omic’ and clinical data have the

potential to expand the ‘predictable’ proportion of future responders within a given RA patient

population, to a point of clinical usefulness and commercial viability. Even as biosimilar anti-TNFα

treatments have reduced in cost, a pharmacogenetic approach to stratify patient populations could

provide a reliable means to rationalise their use toward those most likely to benefit.

Abbreviations

ACPA: anti-citrullinated protein/peptide antibodies; ACR: American College of Rheumatology;

ADCC: antibody-dependent cellular cytotoxicity; ANOVA: Analysis of variance; anti-TNFα: anti-tumour

necrosis factor alpha; BSR: British Society for Rheumatology; cDMARD: conventional Disease-

modifying anti-rheumatic drugs; CCP: cyclic citrullinated peptide; CHUK: conserved helix-loop-helix

ubiquitous kinase; CRP: C-reactive protein; DAS28: Disease activity score assessing 28 joints; ESR:

Erythrocyte sedimentation rate; EULAR: European League Against Rheumatism; HAQ: Health

assessment questionnaire; HLA: human leukocyte antigen; IRAK: interleukin-1 receptor-associated

kinase; KIR: Killer-cell immunoglobulin-like receptor; MAPK: mitogen-activated protein kinase; NFkB:

nuclear factor kappa-light-chain-enhancer of activated B cells; PCR: Polymerase chain reaction;

PTPRC: Protein Tyrosine Phosphatase Receptor Type C; RA: Rheumatoid arthritis; RF: Rheumatoid

factor; SJC: swollen joint count; SNP: Single nucleotide polymorphism; SSOP: sequence-specific

oligonucleotides probes; TJC: tender joint count; TLR: Toll-like receptor; TNF: Tumor necrosis factor.

Acknowledgements

AB and DG wish to acknowledge the award of Ph.D. fellowships from Department for Economy and

support from Northern Ireland Rheumatism Trust, both Northern Ireland, UK. This work was supported

by a grant to AB from Innovate UK and the NI Centre for Stratified Medicine has been funded under

the EU Regional Development Fund EU Sustainable Competitiveness Programme for NI & the NI

Page 12 of 25

Public Health Agency. The datasets used and/or analysed during the current study are available from

the corresponding author on reasonable request.

Declaration of Conflict of Interest

The authors declare that they have no conflict of interest.

Figure and Table Legends

Cohort characteristics  Non-responders (n = 59) 

Moderate responders (n = 87) 

Good responders (n = 92) 

Combined (n=238)

P-value 

Gender, female, n (%)  44 (73.3) 70 (80.5) 66 (71.0) 180 (75.0) 0.32

Age at baseline, years, median (IQR) 

53.6(14.0) 54.7(15.5) 53.1(15.7) 54.5(15.5) 0.94

Disease duration, years 11.5 (10.1) 12.5 (9.9) 11.4 (9.3) 11.8 (9.7) ns

Concurrent cDMARDs, n (%)  51 (85.0) 73 (83.9) 82 (88.2) 206 (85.8%) 0.70

DAS28 score at baseline, mean (s.d.) 

4.9 (1.3) 5.9 (1.1) 5.3 (1.1) 5.4 (1.2) < 0.0001

DAS28 score at outcome, mean (s.d.) 

5.3 (1.2) 4.1 (0.8) 2.1 (0.8) 3.8 (1.6) < 0.0001

Change in DAS28 score, mean (s.d.) 

0.4 (1.0) −1.8 (0.8) −3.1 (1.0) -1.6 (1.6) < 0.0001

baseline TJC, median (IQR)  8.5 (9.8) 12.0 (10.5) 11.0 (10.3) 10.0 (10.0) 0.037

baseline SJC, median (IQR)   5.0 (7.8) 8.0 (7.5) 8.5 (7.3) 8.0 (7.0) 0.0007

baseline CRP, median (IQR)  5.7 (13.4) 13.0 (25.0) 6.7 (14.0) 8.0 (19.7) 0.043

baseline HAQ, median (IQR), n  2.0 (1.0), 45 1.8 (1.0) , 55 1.9 (0.9), 57 1.9 (1.0), 157 0.92

Table 1

Patient characteristics by EULAR response class after six months of anti-TNFα treatment.

Page 13 of 25

Values are mean with standard deviation (s.d.) or interquartile range (IQR), number (n), or percentage

(%), where indicated. DAS28, 28 joint disease activity score; cDMARD, conventional disease

modifying anti-rheumatic drug; HAQ, health assessment questionnaire; CRP, C-reactive protein; SJC,

swollen joint count; TJC, tender joint count; ns, not significant.

Factor Group Group No.s Group % Group Mean delta DAS28

SD p value

Gender Male 59 24.8% -1.70 0.08 0.659

Female 179 75.2% -1.81 -0.03

Baseline cDMARD

No 33 13.9% -1.71 0.08 0.762

  Yes 205 86.1% -1.80 -0.01  

Table 2

Association of baseline clinical factors with change in DAS28 score, over 6 months treatment.

Variable Estimate Std. Error t value Pr(>|t|)(Intercept) 0.957839 0.531244 1.803014 0.0729baseline DAS28 -0.60283 0.084762 -7.11208 2.17 E-11CD226 (12) 0.886626 0.300816 2.947406 0.0036CD226 (22) 0.680279 0.326803 2.081615 0.0386

Table 3

The linear model with two predictors: baseline DAS28 and CD226; After correcting for effects of

baseline DAS28, no other genetic factor or other factors appear to contribute significantly to the delta

DAS28 score; Therefore the two predictors here, baseline DAS28 and CD226 together provide an

optimal model to describe the response of this cohort of RA patients.

Page 14 of 25

Figure 1

Significant factors associated with baseline DAS28 score. A Box plot of baseline DAS28 scores for

RA patients with the HLA-DRB1 0404 Allele absent or present. Central bar represents mean, outer

box standard error and error bars standard deviation of grouping. B Dot plot correlation of baseline

DAS28 scores for RA patients versus baseline health assessment questionnaire score.

Page 15 of 25

Figure 2

Association between genetic and clinical factors and change in DAS28 score after 6 months

treatment. A Association between baseline DAS28 score and delta DAS28. B Association between

baseline cDMARD use and delta DAS28. 0, none used; 1, cDMARD used. C Association between

baseline HAQ score and delta DAS28. D Association between gender and delta DAS28. E

Association between HLA-DRB1 0404 genotypes and delta DAS28. 0, absent; 1, present. F

Association between binary CD226 and delta DAS28. wt, wildtype TT; combined, CT and CC carriers.

Page 16 of 25

Figure 3

Model of combined effects of HLA-DRB1*0404 allele and CD226 SNP rs763361 on change in DAS28

score after 6 months treatment. A Non-interaction plot SNP genotypes: 11, wildtype; 12, heterozygous

carrier; 22 homozygous carrier of CD226 SNP. B Effect of CD226 rs763361SNP presence upon

change in DAS28 ESR score after 6 months of anti-TNFα use; 11 is homozygous wildtype TT, 12 is

CT and 22 is CC genotype. C Combined HLA-DRB1*0404 and binary CD226 effect plot. For HLA-

DRB1*0404 allele, 0 is absent, 1 is present. For CD226 rs763361 SNP, 11 is homozygous wildtype

TT, 12 is CT and 22 is CC genotype.

Page 17 of 25

Supplementary Table Legends

   

RF positiv

e

Anti-CCP

positive

HLA-DRB1 0404

CD226 (TC)

CD226 (CC)

CD226 total

Adalimumab n positive 77 50 25 62 36 98

  n available 100 67 119 119 119 119

  % 77.0% 74.6% 21.0% 52.1% 30.3% 82.4%

Etanercept n positive 47 34 9 32 25 57

  n available 64 44 70 70 70 70

  % 73.4% 77.3% 12.9% 45.7% 35.7% 81.4%

Infliximab n positive 19 7 12 21 7 28

  n available 28 12 35 35 35 35

  % 67.9% 58.3% 34.3% 60.0% 20.0% 80.0%Certoluzimab n positive 5 3 3 6

28

  n available 8 5 8 8 8 8

  % 62.5% 60.0% 37.5% 75.0% 25.0% 100.0%

Golimumab n positive 1 0 1 4 0 4

  n available 2 1 4 4 4 4

  % 50.0% 0.0% 25.0% 100.0% 0.0% 100.0%

Combined n positive 149 94 50 125 70 195

  n available 202 129 236 236 236 236

  % 73.8% 72.9% 21.2% 53.0% 29.7% 82.6%

Supplementary Table 1

Anti-TNFα treatment profile of study cohort. RF, rheumatoid factor; anti-CCP, anti-cyclic citrullinated

peptide; CD226 (TC), heterozygous SNP carrier; CD226 (CC), homozygous SNP carrier; CD226 total,

combined TC and CC carriers.

Factor Group Group No.s Group %Group Mean

Baseline DAS28

p value

Gender Male 47 23.9% 5.16 0.127  Female 150 76.1% 5.47  

Baseline cDMARD No 21 10.7% 5.50 0.694

  Yes 176 89.3% 5.39  KIR2DL2 0 80 40.6% 5.53 0.203

Page 18 of 25

  1 117 59.4% 5.31  KIR2DS2 0 76 38.6% 5.53 0.233

  1 121 61.4% 5.31  HLAC1 0 25 12.7% 5.19 0.362

  1 172 87.3% 5.43  HLAC2 0 86 43.7% 5.58 0.070

  1 111 56.3% 5.26  HLA03 0 155 78.7% 5.42 0.658

  1 42 21.3% 5.32  HLA0101 0 138 70.1% 5.38 0.753

  1 59 29.9% 5.44  HLA0401 0 99 50.3% 5.41 0.907

  1 98 49.7% 5.39  HLA0404 0 155 78.7% 5.30 0.038

  1 42 21.3% 5.75  HLA1001 0 188 95.4% 5.37 0.196

  1 9 4.6% 5.92  HLADRBTNL 11 139 70.6% 5.39 0.932

  12 52 26.4% 5.41    22 6 3.0% 5.58  

PTPRC 11 155 78.7% 5.41 0.353  12 41 20.8% 5.38    22 1 0.5% 3.64  

AFF 11 50 25.4% 5.42 0.981  12 108 54.8% 5.38    22 39 19.8% 5.40  

CD226 11 30 15.2% 5.42 0.921  12 108 54.8% 5.37    22 59 29.9% 5.44  

myD 11 134 68.0% 5.48 0.307  12 58 29.4% 5.25    22 5 2.5% 4.85  

CHUK 11 61 31.0% 5.22 0.403  12 89 45.2% 5.48    22 47 23.9% 5.46  

MTHFR1 11 88 44.7% 5.43 0.928  12 82 41.6% 5.38    22 27 13.7% 5.34  

MTHFR2 11 93 47.2% 5.30 0.567  12 85 43.1% 5.50    22 19 9.6% 5.44  

Supplementary Table 2

Association of factors with baseline DAS28 score. Allele genotypes: 1, present; 0, absent. SNP

genotypes: 11, wildtype; 12, heterozygous carrier; 22 homozygous carrier.

Page 19 of 25

    EULAR RESPONSE, genotype n (%)  

Page 20 of 25

Factor Genotype Group

NON n

% MODERATE n

% GOOD n

% p value

KIR2DL2 0 23 22.3% 42 40.8% 38 36.9% 0.6011 37 27.4% 48 35.6% 50 37.0%

KIR2DS2 0 22 22.2% 40 40.4% 37 37.4% 0.635  1 38 27.3% 50 36.0% 51 36.7%  HLAC1 0 8 27.6% 8 27.6% 13 44.8% 0.463

1 52 24.9% 82 39.2% 75 35.9%HLAC2 0 25 23.4% 44 41.1% 38 35.5% 0.626  1 35 26.7% 46 35.1% 50 38.2%  HLA03 0 45 24.2% 68 36.6% 73 39.2% 0.388

1 15 28.8% 22 42.3% 15 28.8%HLA0101 0 41 24.4% 65 38.7% 62 36.9% 0.877  1 19 27.1% 25 35.7% 26 37.1%  HLA0401 0 33 26.6% 44 35.5% 47 37.9% 0.728

1 27 23.7% 46 40.4% 41 36.0%HLA0404 0 51 27.3% 74 39.6% 62 33.2% 0.060  1 9 17.6% 16 31.4% 26 51.0%  HLA1001 0 57 24.9% 87 38.0% 85 37.1% 0.849  1 3 33.3% 3 33.3% 3 33.3%  HLADRBTNL 11 40 23.5% 62 36.5% 68 40.0% 0.488

12 18 29.5% 24 39.3% 19 31.1%22 2 28.6% 4 57.1% 1 14.3%

PTPRC 11 41 22.2% 75 40.5% 69 37.3% 0.18812 19 37.3% 14 27.5% 18 35.3%

  22 0 0.0% 1 50.0% 1 50.0%  AFF 11 41 22.2% 75 40.5% 69 37.3% 0.886

12 19 37.3% 14 27.5% 18 35.3%22 0 0.0% 1 50.0% 1 50.0%

CD226 11 7 17.1% 13 31.7% 21 51.2% 0.20212 36 28.6% 51 40.5% 39 31.0%

  22 17 23.9% 26 36.6% 28 39.4%  myD88 11 44 27.3% 64 39.8% 53 32.9% 0.220

12 13 18.6% 24 34.3% 33 47.1%22 3 42.9% 2 28.6% 2 28.6%

CHUK 11 22 27.8% 31 39.2% 26 32.9% 0.83412 23 22.1% 40 38.5% 41 39.4%

  22 15 27.3% 19 34.5% 21 38.2%  MTHFR1 11 27 25.0% 47 43.5% 34 31.5% 0.045

12 22 22.2% 30 30.3% 47 47.5%22 11 35.5% 13 41.9% 7 22.6%

MTHFR2 11 32 27.4% 44 37.6% 41 35.0% 0.74512 21 21.6% 36 37.1% 40 41.2%

  22 7 29.2% 10 41.7% 7 29.2%  

Supplementary Table 3

Association of genetic factors with EULAR response.

Factor Group Group No.s Group % Group Mean delta DAS28

p value

KIR2DL2 0 103 43.3% -1.89 0.398

Page 21 of 25

1 135 56.7% -1.71KIR2DS2 0 99 41.6% -1.91 0.320  1 139 58.4% -1.70  HLAC1 0 29 12.2% -1.76 0.929

1 209 87.8% -1.79HLAC2 0 107 45.0% -1.84 0.684  1 131 55.0% -1.75  HLA03 0 186 78.2% -1.86 0.178

1 52 21.8% -1.51HLA0101 0 168 70.6% -1.76 0.708  1 70 29.4% -1.85  HLA0401 0 124 52.1% -1.83 0.653

1 114 47.9% -1.74HLA0404 0 187 78.6% -1.67 0.033  1 51 21.4% -2.22  HLA1001 0 229 96.2% -1.80 0.489

1 9 3.8% -1.41HLADRBTNL 11 170 86.3% -1.89 0.329

12 61 31.0% -1.55  22 7 3.6% -1.42  PTPRC 11 185 93.9% -1.85 0.281

12 51 25.9% -1.5322 2 1.0% -3.00

AFF 11 59 29.9% -1.61 0.58812 126 64.0% -1.81

  22 53 26.9% -1.93  CD226 11 41 20.8% -2.35 0.029

12 126 64.0% -1.5722 71 36.0% -1.85

myD88 11 161 81.7% -1.77 0.91612 70 35.5% -1.84

  22 7 3.6% -1.61  CHUK 11 79 40.1% -1.65 0.479

12 104 52.8% -1.9322 55 27.9% -1.71

MTHFR1 11 108 54.8% -1.71 0.11612 99 50.3% -2.01

  22 31 15.7% -1.34  MTHFR2 11 117 59.4% -1.67 0.341

12 97 49.2% -1.97  22 24 12.2% -1.58  

Supplementary Table 4

Association of genetic factors with change in DAS28 score over 6 months treatment.

Variable Estimate Std. Error t value Pr(>|t|)(Intercept) 2.48595 0.680512 3.653056 0.000358Baseline DAS28 -0.66396 0.110541 -6.0065 1.39E-08Concurrent cDMARD -0.76153 0.386531 -1.97016 0.050674

Page 22 of 25

Baseline HAQ 0.10124 0.157878 0.641258 0.52234Gender 0.004174 0.301256 0.013856 0.988964

Supplementary Table 5

The linear model with four potential predictors.

Responder Non-responder

Biomarker Prevalence TP FN TN FP Sensitivity TP/(TP+FN)

Specificity TN/(TN+FP)

NPV TN/(FN+TN)

PPV TP/(TP+FP)

FDR 1-PPV

HLA-DRB1*0404 21% 42 139 52 9 23% 85% 27% 82% 18%

CD226 rs763361 17% 34 146 54 7 19% 89% 27% 83% 17%

Supplementary Table 6

Sensitivity and specificity of two genetic predictors of anti-TNFα response. TP, true positive; FN, false

negative; TN, true negative; FP, false positive; NPV, negative predictive value; PPV, positive

predictive value; FDR, false discovery rate.

References

1. Hyrich KL, Watson KD, Silman AJ, Symmons DP, British Society for Rheumatology Biologics Register: Predictors of response to anti-TNF-alpha therapy among patients with rheumatoid

Page 23 of 25

arthritis: results from the British Society for Rheumatology Biologics Register. Rheumatology (Oxford) 2006; 45: 1558-65.

2. Stevenson M, Archer R, Tosh J et al.: Adalimumab, etanercept, infliximab, certolizumab pegol, golimumab, tocilizumab and abatacept for the treatment of rheumatoid arthritis not previously treated with disease-modifying antirheumatic drugs and after the failure of conventional disease-modifying antirheumatic drugs only: systematic review and economic evaluation. Health Technol.Assess. 2016; 20: 1-610.

3. Kleinert S, Tony HP, Krause A et al.: Impact of patient and disease characteristics on therapeutic success during adalimumab treatment of patients with rheumatoid arthritis: data from a German noninterventional observational study. Rheumatol.Int. 2012; 32: 2759-67.

4. Hetland ML, Stengaard-Pedersen K, Junker P et al.: Radiographic progression and remission rates in early rheumatoid arthritis - MRI bone oedema and anti-CCP predicted radiographic progression in the 5-year extension of the double-blind randomised CIMESTRA trial. Ann.Rheum.Dis. 2010; 69: 1789-95.

5. Klaasen R, Cantaert T, Wijbrandts CA et al.: The value of rheumatoid factor and anti-citrullinated protein antibodies as predictors of response to infliximab in rheumatoid arthritis: an exploratory study. Rheumatology (Oxford) 2011; 50: 1487-93.

6. Alessandri C, Bombardieri M, Papa N et al.: Decrease of anti-cyclic citrullinated peptide antibodies and rheumatoid factor following anti-TNFalpha therapy (infliximab) in rheumatoid arthritis is associated with clinical improvement. Ann.Rheum.Dis. 2004; 63: 1218-21.

7. Potter C, Cordell HJ, Barton A et al.: Association between anti-tumour necrosis factor treatment response and genetic variants within the TLR and NF{kappa}B signalling pathways. Ann.Rheum.Dis. 2010; 69: 1315-20.

8. Nair SC, Welsing PM, Choi IY et al.: A Personalized Approach to Biological Therapy Using Prediction of Clinical Response Based on MRP8/14 Serum Complex Levels in Rheumatoid Arthritis Patients. PLoS One 2016; 11: e0152362.

9. Sode J, Vogel U, Bank S et al.: Confirmation of an IRAK3 polymorphism as a genetic marker predicting response to anti-TNF treatment in rheumatoid arthritis. Pharmacogenomics J. 2016.

10. Bek S, Bojesen AB, Nielsen JV et al.: Systematic review and meta-analysis: pharmacogenetics of anti-TNF treatment response in rheumatoid arthritis. Pharmacogenomics J. 2017; 17: 403-11.

11. Zervou MI, Myrthianou E, Flouri I et al.: Lack of association of variants previously associated with anti-TNF medication response in rheumatoid arthritis patients: results from a homogeneous Greek population. PLoS One 2013; 8: e74375.

12. Coulthard LR, Taylor JC, Eyre S et al.: Genetic variants within the MAP kinase signalling network and anti-TNF treatment response in rheumatoid arthritis patients. Ann.Rheum.Dis. 2011; 70: 98-103.

13. Potter C, Hyrich KL, Tracey A et al.: Association of rheumatoid factor and anti-cyclic citrullinated peptide positivity, but not carriage of shared epitope or PTPN22 susceptibility variants, with anti-tumour necrosis factor response in rheumatoid arthritis. Ann.Rheum.Dis. 2009; 68: 69-74.

14. Jiang X, Askling J, Saevarsdottir S et al.: A genetic risk score composed of rheumatoid arthritis risk alleles, HLA-DRB1 haplotypes, and response to TNFi therapy - results from a Swedish cohort study. Arthritis Res.Ther. 2016; 18: 288-z.

15. Skapenko A, Smolen JS, Kavanaugh A, Arora V, Kupper H, Schulze-Koops H: Genetic markers associated with clinical and radiographic response in adalimumab plus methotrexate- or methotrexate-treated rheumatoid arthritis patients in OPTIMA. Clin.Exp.Rheumatol. 2019; 37: 783-90.

16. Viatte S, Barton A: Genetics of rheumatoid arthritis susceptibility, severity, and treatment response. Semin.Immunopathol. 2017; 39: 395-408.

17. McGeough CM, Berrar D, Wright G et al.: Killer immunoglobulin-like receptor and human leukocyte antigen-C genotypes in rheumatoid arthritis primary responders and non-responders to anti-TNF-alpha therapy. Rheumatol.Int. 2012; 32: 1647-53.

18. Arnett FC, Edworthy SM, Bloch DA et al.: The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum. 1988; 31: 315-24.

19. Prevoo ML, van 't Hof, M. A., Kuper HH, van Leeuwen MA, van de Putte, L. B., van Riel PL: Modified disease activity scores that include twenty-eight-joint counts. Development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum. 1995; 38: 44-8.

Page 24 of 25

20. van Gestel AM, Prevoo ML, van 't Hof, M. A., van Rijswijk MH, van de Putte, L. B., van Riel PL: Development and validation of the European League Against Rheumatism response criteria for rheumatoid arthritis. Comparison with the preliminary American College of Rheumatology and the World Health Organization/International League Against Rheumatism Criteria. Arthritis Rheum. 1996; 39: 34-40.

21. Cui J, Taylor KE, Destefano AL et al.: Genome-wide association study of determinants of anti-cyclic citrullinated peptide antibody titer in adults with rheumatoid arthritis. Mol.Med. 2009; 15: 136-43.

22. Cui J, Saevarsdottir S, Thomson B et al.: Rheumatoid arthritis risk allele PTPRC is also associated with response to anti-tumor necrosis factor alpha therapy. Arthritis Rheum. 2010; 62: 1849-61.

23. Tan RJ, Gibbons LJ, Potter C et al.: Investigation of rheumatoid arthritis susceptibility genes identifies association of AFF3 and CD226 variants with response to anti-tumour necrosis factor treatment. Ann.Rheum.Dis. 2010; 69: 1029-35.

24. Xiao H, Xu J, Zhou X et al.: Associations between the genetic polymorphisms of MTHFR and outcomes of methotrexate treatment in rheumatoid arthritis. Clin.Exp.Rheumatol. 2010; 28: 728-33.

25. Owen SA, Lunt M, Bowes J et al.: MTHFR gene polymorphisms and outcome of methotrexate treatment in patients with rheumatoid arthritis: analysis of key polymorphisms and meta-analysis of C677T and A1298C polymorphisms. Pharmacogenomics J. 2013; 13: 137-47.

26. MacGregor A, Ollier W, Thomson W, Jawaheer D, Silman A: HLA-DRB1*0401/0404 genotype and rheumatoid arthritis: increased association in men, young age at onset, and disease severity. J.Rheumatol. 1995; 22: 1032-6.

27. Meyer JM, Evans TI, Small RE et al.: HLA-DRB1 genotype influences risk for and severity of rheumatoid arthritis. J.Rheumatol. 1999; 26: 1024-34.

28. Criswell LA, Lum RF, Turner KN et al.: The influence of genetic variation in the HLA-DRB1 and LTA-TNF regions on the response to treatment of early rheumatoid arthritis with methotrexate or etanercept. Arthritis Rheum. 2004; 50: 2750-6.

29. Guis S, Balandraud N, Bouvenot J et al.: Influence of -308 A/G polymorphism in the tumor necrosis factor alpha gene on etanercept treatment in rheumatoid arthritis. Arthritis Rheum. 2007; 57: 1426-30.

30. Kang CP, Lee KW, Yoo DH, Kang C, Bae SC: The influence of a polymorphism at position -857 of the tumour necrosis factor alpha gene on clinical response to etanercept therapy in rheumatoid arthritis. Rheumatology (Oxford) 2005; 44: 547-52.

31. Murdaca G, Spano F, Contatore M, Guastalla A, Magnani O, Puppo F: Pharmacogenetics of etanercept: role of TNF-alpha gene polymorphisms in improving its efficacy. Expert Opin.Drug Metab.Toxicol. 2014; 10: 1703-10.

32. Boyton RJ, Altmann DM: Natural killer cells, killer immunoglobulin-like receptors and human leucocyte antigen class I in disease. Clin.Exp.Immunol. 2007; 149: 1-8.

33. Fasth AE, Bjorkstrom NK, Anthoni M, Malmberg KJ, Malmstrom V: Activating NK-cell receptors co-stimulate CD4(+)CD28(-) T cells in patients with rheumatoid arthritis. Eur.J.Immunol. 2010; 40: 378-87.

34. Qiu ZX, Zhang K, Qiu XS, Zhou M, Li WM: CD226 Gly307Ser association with multiple autoimmune diseases: a meta-analysis. Hum.Immunol. 2013; 74: 249-55.

35. Shirakawa J, Shibuya K, Shibuya A. Requirement of the serine at residue 329 for lipid raft recruitment of DNAM-1 (CD226). Int Immunol. 2005; 3: 217-23.

 

Page 25 of 25