Precision medicine and drug optimization in adult inflammatory bowel disease patients

Genomics

Genetic polymorphism could influence the response to treatment by a series of mechanisms including drug clearance or immunogenicity and have the advantage of not changing over time. There are therefore several reasons to believe that SNP could be potential biomarkers to predict response to treatment, maybe to identify slow responders from early complete responders and to help selecting the best therapy for each individual patient.^139,202 In a systematic review including 15 studies, Bek et al. ¹³⁹ attempted to identify potential genetic markers to predict anti-TNF response in UC and CD patients. They highlighted that genes involved in the innate immune response such as recognition of bacterial components and cytokine pathways could be important for the response to anti-TNF. ¹³⁹ Indeed, toll-like receptor (TLR) 4 (rs55030728), FC fragments of IgG receptor IIIa (FCGR3A, rs396991), tumour necrosis factor receptor superfamily 1A (TNFRS1A, rs4149570), interferon-gamma (IFNγ, rs2439561), IL-6 (IL-6, rs10499563) and interleukin-1B (IL-1B, rs4848306) polymorphism were associated with improved anti-TNF response in IBD. ¹³⁹ TLR2 (rs3804099) and TLR9 (rs352139) SNPs were, on the contrary, associated with a poorer response. ¹³⁹ A number of genes among the 200 susceptibility genes identified in IBD population were also studied to see whether it could predict the response to biological.^{104,250–252} Although being the strongest susceptibility gene identified for CD, ⁸⁹ NOD2/CARD15 polymorphism was not found to be a good predictor of response to infliximab or adalimumab in CD patients.^104,253 In contrast, variants in IL-23R, another IBD susceptibility gene, seem to be associated with the response to anti-TNF. ¹³⁰ Homozygous carriers of IBD risk-increasing IL23R variants (rs1004819, rs2201841, rs10889677, rs11209032 and rs1495965) were more likely to respond to infliximab therapy than homozygous carriers of IBD risk-decreasing IL23R variants (rs7517847, rs10489629, rs11465804 and rs1343151). ¹³⁰ IL-23 receptors are highly expressed on the surface of Th17 cells and are important for the differentiation of these cells producing TNF-α, which could explain links between IL23R genotype and response to anti-TNF. ¹³⁰ However, only a minority of IBD patients are homozygous for these IL23R variants, making the use of IL23R genotyping for predicting response to infliximab limited in clinical practice. ¹³⁰ Hlavaty et al. ²⁰⁴ have also reported that polymorphisms in apoptosis genes (Fas ligand and caspase-9 gene) could influence response to infliximab in luminal and fistulizing CD. Patients with a Fas ligand −843 CC/CT genotype and caspase-9 93 TT genotype have a higher rate of clinical response to infliximab than those with Fas ligand −843 TT genotype and caspase-9 93 CC and CT genotype, respectively. ²⁰⁴ Unfortunately, these findings are not systematically found from one study to another. It was indeed demonstrated on a prospective cohort of 121 CD patients that patients with a Fas ligand CC genotype being a non-responder were fourfold higher as compared to a TC or TT genotype ²⁵⁴

While some studies have shown that models combining genetic and clinical variables were superior to a model including only clinical variables to predict PNR to anti-TNF,^21,202 other studies found that the addition of genetic markers did not provide any benefit to these predictive models. ¹⁹⁰ To date, few and weak genetic biomarkers were identified to predict response to treatment. These are sometimes not reproducible and are associated with polymorphisms rarely present in the population, which makes their use in clinical practice limited. ¹³⁹ Furthermore, the expression of these potential genes could be influenced by environmental factors (or the exposome). These can act by modulating the epigenome, which are poorly considered in studies looking for genes associated with treatment response, and deserved to be taken into consideration. ²⁵⁵ The genome sequencing for SNP identification in IBD patients is not currently considered to bring sufficient benefit to justify the associated costs. ²⁵⁶

Transcriptomics

If the study of genomics has not allowed to find sufficiently robust genetic markers to predict response to anti-TNF to date, transcriptomics is another science that could bring complementary data. Transcriptomic studies can be performed on intestinal mucosa biopsies or on blood samples. By analysing the mRNA expression profiles of colonic mucosal biopsies from UC patients enrolled in the Active Ulcerative Colitis Trial 1 (ACT1), Toedter et al. found an expression difference of genes involved in the Th1, Th2 and Th17 pathways between responders and non-responders to infliximab. ¹⁵¹ Similarly, Arijs et al. ¹⁴² compared the pre-treatment colonic mucosal biopsy-derived mRNA expression between responders and non-responders to provide a predictive response signature for infliximab treatment in UC. They found that the downregulation of a combination of five genes (TNFRSF11B, STC1, PTGS2, IL13Ralpha2 and IL11) predicted the response to infliximab at weeks 4–8 with 89% of accuracy. ¹⁴² A few years later, still in UC patients, but for golimumab this time, baseline gene expression signature was studied to see whether it could be used to predict patients which would achieve mucosal healing, clinical response and clinical remission at weeks 6 and 30 of this anti-TNF. ¹⁴⁷ A 13-gene signature (CMTM2, C5AR1, FGF2, GK, HGF, IL1RN, LILRA2, NAMPT, PAPPA, SNCA, SOD2, STEAP4, ZBED3) was also highlighted to predict mucosal healing at week 6 (note that this was not significant in predicting clinical remission or response). ¹⁴⁷ For CD patients, the same group (Arijs et al. and colleagues) therefore investigated whether they could identify a mucosal gene panel allowing reliable prediction of response to infliximab. The top five differentially expressed genes at baseline (TNFAIP6, S100A8, IL11, G0S2 and S100A9) allowed to distinguish responders from non-responders to anti-TNF with an overall accuracy of 100% in patients with colonic involvement. ²²⁵ Although these studies demonstrated that transcriptional profiles could be useful to predict anti-TNF response, no overlap was found between the identified genes. ²³ Finally, a study performed on 48 UC patients showed that non-responders have a more severe proinflammatory cytokine profile with higher IL-1β, IL-17A, IL-6 and IFN-γ mucosal mRNA expression, ¹⁴⁶ but these results were not unanimous and other studies reported that higher gene expression levels of IL-17A and IFN-γ were significantly associated with remission after infliximab. ¹⁵²

Because of their identified role in the IBD pathophysiology, the role of more specific biomarkers has been investigated. The oncostatin M (OSM), a cytokine belonging to the IL-6 cytokine family, whose expression is increased in inflammatory intestinal tissues, would seem to be an inflammatory amplifier and driver of disease chronicity by promoting chemokines, cytokines and adhesion-factor production by intestinal stromal cells. ¹⁴³ It was demonstrated on an analysis including two cohorts from phase III clinical trials of infliximab and golimumab (more than 200 UC and CD patients), that high baseline expression of OSM was strongly associated with anti-TNF failure. ¹⁴³ Other mucosal biomarkers have also been suggested but with less robust data. The downregulation of the T-cell activation RhoGTPase activating protein or of the triggering receptor expressed on myeloid cells 1 (TREM1) mRNA expression levels at baseline was indeed associated with better chance of responding to anti-TNF.^149,151 The upregulation of this latter marker (TREM1) in whole blood was also found to be rather predictive of non-response to anti-TNF, as well as the upregulation of chemokine receptor type 2 (CCR2)-chemokine ligand 7 (CCL7).^149,150 Peripheral blood markers have the advantage of being more easily accessible and associated with less patient discomfort, making them very attractive targets for the future. ²³

Although less widely studied than to predict response to anti-TNF, mucosal gene expression has also been used to attempt to predict response to anti-integrins or to anti-IL12/IL-23. Similar to what they did to try to identify genes that predict response to infliximab in patients with UC and CD, Arijs et al. analysed mucosal gene expression in UC patients treated by vedolizumab. However, no gene has been identified as predictive of vedolizumab response by comparing the pre-treatment array profiles of responders with non-responders. ¹⁴⁵ In contrast, Verstock et al. identified, in a cohort of 31 IBD patients, four genes (RGS13, DCHS2, MAATS1 and PIWIL1) whose baseline expression levels in colon tissues could predict endoscopic remission with vedolizumab. ¹⁷⁶ Preliminary data were also available for etrolizumab, another anti-integrin targeting the β₇ subunit of the heterodimeric integrins α₄β₇ and α_Eβ₇, before the development was stopped. ²⁵⁷ The presence of increase levels of granzyme a (GZMA) and integrin αE gene mRNAs in colon tissues of UC patients could have allowed IBD specialists to identify patients who are more likely to respond to this treatment.^258,259 Finally, for ustekinumab, Nishioka et al. ¹⁴⁴ have recently demonstrated that higher mucosal IL23A expression predicts the response to ustekinumab in both UC and CD patients.

History has unfortunately shown that biomarkers found in one cohort were not necessarily found in another. ¹⁴⁷ They must also be interpreted with caution because some are simply associated with response to treatment but are not necessarily predictive of it (a remark also applicable to other types of markers). ⁵⁶ To date, all these highlighted biomarkers need therefore to be validated on independent cohorts before being implemented in clinical practice.

Proteomics and proteins expression

Transcriptomics does not take into account post-translational modifications and does not always allow to have information on the functional repercussions of the mRNA observed changes. ²³ To this end, the proteomic or other tools of protein quantification, including immunohistochemistry and flow cytometry, could give more precise indications on how cell function could be impacted and with it, the response to different IBD treatments. ²³

A proteomic analysis performed on mucosal biopsies of 56 biologic-naïve UC patients showed that anti-TNF responders and non-responders had differential pattern expressions of anti-microbial peptide AMP and cytokines. ¹⁵⁴ Patients responders have higher levels of defensin-5α, eosinophil cationic protein in mucosa, whereas non-responders have higher levels of cathelicidin antimicrobial peptide, IL-12, IL-17A at baseline. ¹⁵⁴ Proteomic analyses can also be carried on the blood. A study performed in 47 infliximab-naïve IBD patients showed that patients with a response to infliximab at week 14 had higher transmembrane TNF-α (tmTNF-α) in their circulating lymphocytes and monocytes. ¹⁵⁵ Studies are not unanimous regarding the role of TNF-α serum levels in predicting response to anti-TNF. While some have suggested that a higher basal level of TNF-α may be predictive of response to infliximab in UC patients, ¹⁵⁸ another study demonstrated that a higher level of serum TNF-α before treatment initiation was associated with lack of response to anti-TNF. ²³⁰ Locally in the mucosa, high mucosal TNF-α levels before treatment might instead be associated with a better response to anti-TNF in CD patients. ²⁶⁰ This discrepancy can be explained by the fact that, in humans, the abundance of proteins (such as cytokines) is subject to large genetic variations and that individual differences in abundance do not always reflect individual differences in biological activity. ¹⁵⁸ Indeed, for the same level of cytokines, the activity of these may be different and the underlying cytokines composition may also differ. ¹⁵⁸ It may therefore be more interesting to study a coordinated (matrix-evaluated) cytokine response rather than the level of a single cytokine to predict response to treatment. ¹⁵⁸ Using seven cytokines (TNF-α, IL-12, IL-8, IL-2, IL-5, IL1-β and IFN-γ), Obraztsov et al. ¹⁵⁸ proposed a model allowing to classify patients into responders and non-responders with a sensitivity of 84.2% and a specificity of 93.3%. Increases in other cytokines were also associated with a lack of response, but with the same limitation. It was thus demonstrated in 20 CD patients that those not responding to anti-TNF had a higher frequency of Treg (assessed by flow cytometry) in pre-treatment and transforming growth factor (TGF)−β1 level [assessed by enzyme-linked immunosorbent assay (ELISA)]. ²²⁸

Some teams have also tried to use proteomic or proteins expression data to try to predict the response to IBD therapies other than anti-TNFs. By analysing immunophenotyping of peripheral blood mononuclear cells and expression of α₄β₇ integrin on lymphocytes from 26 IBD patients, Boden et al. ¹⁷⁷ sought to identify biomarkers associated with response to vedolizumab. They found that α₄β₇ expression on multiple subsets of T, B and natural killer (NK) cells was higher in responders than in non-responders prior to treatment administration. ¹⁷⁷ Another study using multiplex ELISA to quantified 47 preselected plasma proteins in blood samples of 28 anti-TNF refractory IBD patients prior to initiation of vedolizumab demonstrated that higher IL-6 and soluble CD-40 ligand circulating level could predict non-response to vedolizumab (only in UC patients for CD-40 ligand) while higher levels of osteocalcin could predict response. ¹⁷⁸ Osteocalcin (a marker of bone formation), although increased in responders compared to non-responders, appears to be of limited value in predicting response to vedolizumab and guiding treatment choice. Indeed, the high level of bone formation may only reflect relative bone sparing in patients with less severe disease or less long-standing inflammation. ¹⁷⁸ In contrast, IL-6 levels may be more informative. As previously mentioned, high levels of IL-6 have also been described in patients refractory to anti-TNF drugs, suggesting that IL-6 may play a role in non-TNF mediated inflammation.^146,178,261 These data suggest that patients with higher IL-6 signalling would benefit from therapy other than anti-TNF or vedolizumab.^145,178 For ustekinumab, Creyns et al. ²⁶² demonstrated, in 46 anti-TNF and vedolizumab refractory CD patients, that responders had lower baseline level of innate lymphoid cells expressing transcription factors/cytokines present in Th1 cells (also known as ILC1s cells) while non-responders had higher rates. Finally, a series of anti-IL23 drugs will probably reach the market in the near future. ¹⁰³ For these therapies, baseline serum concentration of IL-22 (an upstream regulator of IL-23) could be used to predict response to treatment at week 8. Indeed, patients with an IL-22 concentration greater than or equal to 15.6 pg/mL were more likely to respond to anti-IL23p19. ²⁶³

Similar to what was observed with transcriptomics, these studies describing potential biomarkers are encouraging but have generally been observed in small cohorts and need to be validated in larger cohorts before being applied in clinical practice. ²³

Metabolomics

Metabolomics could also help in predicting the response to biotherapies. ³⁶ In a prospective longitudinal cohort study enrolling 76 CD patients, Ding et al. ²⁶⁴ demonstrated that a range of metabolic biomarkers may contribute to prediction of response to anti-TNF therapy in CD patients. Primary non-responders tended to have modification of serum lipids (higher level of ceramide and sphingomyelin) as well as serum and faecal bile acids changes (higher levels of circulating primary unconjugated bile acids as well as higher levels of bile acids conjugated to sulphate, taurine and glycine in faeces). ²⁶⁴ In contrast, anti-TNF responders had a higher histidine levels in faeces and serum as well as a higher urinary cysteine level and the ROC analysis demonstrated a good model for predicting anti-TNF response for this latter. ²⁶⁴ Faecal metabolites’ profiles could also provide some answers. For example, butyrate and substrates involved in butyrate synthesis (e.g. acetaldehyde) have been identified as predictive metabolites of clinical remission following biologic therapy.^36,265 The presence of these metabolites in the stool is influenced by the microbiota, which can, of course, also be used in precision medicine.

Radiomics

A few studies have evaluated the ability of imaging-based biomarkers to predict, at diagnosis, the response to specific treatments. Rimola et al. ²³⁵ showed that CD patients with creeping fat or ileal lesions at pre-treatment MRE were unlikely to heal severe inflammation on a long term under anti-TNFα treatment. Some criteria assessed in MRE could also help predict which CD patients with symptomatic ileal stenosis are likely to respond to anti-TNF therapy. ²³³ In the CREOLE study, CD patients with four criteria or more (including MRE ones) among the following (the use of an immunomodulator, the presence of obstructive symptoms for <5 weeks, a Crohn disease obstructive score >4, a small bowel stricture length of <12 cm, a maximal small bowel diameter proximal to stricture(s) of 18–29 mm, a marked enhancement on delayed phase and absence of a fistula) were considered to be likely to respond to adalimumab. ²³³ The presence of an apparent diffusion coefficient < 1 × 10⁻³ mm²/s assessed by diffusion-weighted MRE could also predict response to anti-TNF in CD patients with stricture. ²³⁴ A team also looked at the ability of visceral adipose tissue volume and visceral fat index (visceral: subcutaneous adipose tissue ratio; VFI), assessed by computed tomography scans, to predict clinical response and C-reactive protein (CRP) reduction in response to anti-TNFα initiation. ¹⁶¹ IBD patients with visceral adipose tissue volume of 1500–2999 cm³ were most likely to response and to have a mean CRP reduction at 12 months (compared with those with a volume ⩾3000 cm³). ¹⁶¹ In contrast, patients with VFI ⩾ 0.67 were significantly more likely to undergo surgery at 6 and 12 months compared with those with VFI < 0.33. ¹⁶¹ Finally, nomograms based on radiomics were established to predict mucosal healing or SLR with infliximab in CD patients.^236–238 Zhu et al. ²³⁷ indeed attempted to predict mucosal healing in biologic-naïve CD patients treated with infliximab using clinical factors and radiomics features. Their clinical radiomics nomogram combined disease duration and a computed tomography enterography (CTE)-based radiomics signature at baseline and performed well to predict mucosal healing in CD patients after 26 weeks of infliximab treatment. ²³⁷ Feng et al. ²³⁶ also established and validated a nomogram based on an MRI-based radiomic index able to detect change in iron metabolism to identify CD patients at risk of SLR to infliximab. Chen et al. ²³⁸ also developed a radiomic nomogram with eight radiographic features to predict loss of response to infliximab in CD patients.

Microbiome

Playing a key role in the initiation and propagation of intestinal inflammation in IBD, and modifying in response to IBD treatments, the gut microbiome could also be used to predict attenuation of inflammation in response to biologic treatments.^266–270 While for anti-TNF, the presence of a more diverse microbiome at baseline was not predictive of treatment response,^269,271 it was, however, associated with clinical remission with vedolizumab (and even predictive for this treatment) and ustekinumab.^180,185 More specifically, it has been shown by several studies that disturbances in the taxa that typically produce short-chain fatty acids (SCFA) were associated with conventional and biologics treatment failures.^272,273 Indeed, studies have shown that patients more likely to achieve clinical remission with anti-TNF seem to have greater abundance of species producing butyrate prior treatment initiation such as Clostridium citroneae or Agathobaculum butyricproduces.^266,274 Consistent with these data, it has been demonstrated that the abundance of another SCFAs producers, F. prausnitzii, tended to be higher in anti-TNF responders than in non-responders at baseline, ¹⁵⁴ but this association was not found in all studies. ²⁶⁶ The same is true for patients treated by vedolizumab, for whom, a greater abundance of other SCFAs producers species at baseline, Roseburia inulinivorans and Burkholderiales, was also predictive of remission at week 14. ¹⁸⁰ The same authors highlighted the importance of predictive models incorporating both clinical and microbiome data (such as vedoNet, a network algorithm) in predicting clinical remission. ¹⁸⁰ Finally, a study investigated the association between faecal microbiota composition and response to ustekinumab in 232 anti-TNF refractory CD patients (from phase 2 CERTIFI study). ¹⁸⁵ Faecalibacterium and Bacteroides species were significantly more abundant at baseline in subjects who were in remission 6 weeks after ustekinumab treatment than those who were not, ¹⁸⁵ while Escherichia or Shigella were lower. ¹⁸⁵

The use of microbiota as predictors of response is made difficult by the fact that the identified disturbance in microbiota may be the reflect of the inflammatory burden or any other confounding factors, including environmental ones. ²¹ For some, microbiota may be a better biomarker for predicting response to gut-specific therapies, such as vedolizumab, than systemic therapies such as anti-TNF. ¹⁸⁰ Lee et al. ²⁶⁶ pointed that the association between microbiome and clinical or endoscopic outcomes was stronger for week 14 than for week 52, suggesting that microbiome may have a greater impact on short-term outcomes. Studies systematically comparing the evolution of microbial profiling with the clinical course in response to a specific treatment would help identify the optimal prediction window of the microbiome for individual clinical outcomes. ²⁶⁶

Others and visualization tools

The use of confocal laser endomicroscopy with the topically application of fluorescent antibodies (targeting anti-TNF or α₄β₇ integrin) directly onto the disease mucosae allows the detection and quantification of mTNF-bearing or α₄β₇-positive mucosal cells and to predict response to anti-TNF and to vedolizumab, respectively.^240,245 These findings need to be validated in independent larger multicentre cohorts but the use of these fluorescent antibodies in relapsing patients for whom an endoscopy is performed prior to a treatment change would allow us to measure the amount of molecules that is targeted by a particular therapy and could guide the therapeutic strategy. ²¹ In 2019, Martin et al. ²⁴¹ showed, using single-cell analysis of ileal inflamed tissues from CD patients, that the presence of GIMAT module at diagnosis was correlated with failure to achieve durable CS-free remission upon anti-TNF therapy. The GIMAT organization refers to a unique cellular module with IgG plasma cells, inflammatory mononuclear phagocytes, activated T cells and stromal cells and is driven by a unique MNP-dependent cytokine/chemokine network. ²⁴¹

Some have created clinical decision support tools (CDST) to predict outcomes of treatment with vedolizumab in UC and CD patients. ¹⁸¹ In UC patients, using the absence of exposure to an anti-TNF (+3 points), disease duration of 2 years or more (+3 points), baseline endoscopic activity (moderate versus severe) (+2 points) and baseline albumin concentration (+0.65 points per 1 g/L), they could determine, on a validation cohort, with a high sensitivity (93%), that patients with a score of 26 points or less did not respond to vedolizumab. ¹⁸¹ Patients with a score of 27–32 points or 33 points or more had intermediate and high probability of vedolizumab response (corticosteroid-free remission at week 26), respectively. ¹⁸¹ Similarly, a scoring system has been developed and validated to identify CD patients most likely to respond to 26 weeks of vedolizumab treatment. ¹⁷³ The following factors were used: previous anti-TNF treatment (+3 points), absence of prior bowel surgery (+2 points), absence of prior fistulizing disease (+2 points), baseline level of albumin (+0.4 points per g/L) and baseline concentration of CRP (reduction of 0.5 points for values between 3.0 and 10.0 mg/L and 3.0 points for values >10.0 mg/L). ¹⁷³ The cut-off value of ⩽13 points allowed us to identify, with a good sensitivity, CD patients most likely to respond. ¹⁷³

Other visualization tools, such as nomogram, or machine learning related to artificial intelligence, have also been applied to guide drug use in IBD. Nomogram is a graphical representation of a mathematical formula (or of an algorithm) in which are incorporated several predictors models as continuous variables to predict an end point and facilitates patient management-related decisions by providing superior individualized disease-related risk estimations. ²⁷⁵ These are based on traditional statistical methods, such as multivariable logistic regression and Cox proportional hazards analysis. ²⁷⁵ Nomograms based on radiomics have been presented above. Nomograms were developed, from IBD Bioresourse (United Kingdom), to predict surgery for IBD patients initially treated with methotrexate in monotherapy. ¹²⁷ For UC, sex and age at diagnosis were predictors in the nomogram. ¹²⁷ For CD, gender, treatment era, tolerance, lesion site, perianal involvement, disease behaviour and biologics requirements were in the nomogram. ¹²⁷ Using multiple logistic regression, Chen et al. ²⁴² found that disease behaviour, body mass index (BMI), CRP and IL-6 levels before infliximab initiation were predictive factors of PNR to infliximab at week 14 in CD patients and developed and validated an IL-6 nomogram. Finally, a nomogram based on bioelectrical impedance analysis indexes (based on body composition parameters) and laboratory markers (haemoglobin, albumin, serum iron) could predict SLR to infliximab in bio-naïve CD patients at 54 weeks. ²³⁸ Other nomograms have been proposed to predict response to treatment but were based on factors other than those present at baseline. ²⁷⁶

Machine learning belongs to the field of artificial intelligence and refers to the ability of computers to learn to make decisions or detect patterns from data, without explicitly being programmed. ²⁷⁷ Several machine learning predictive models have already been suggested such as (1) a machine learning algorithm to predict clinical remission with thiopurines ²⁷⁸ ; (2) a machine learning to predict non-durable response to anti-TNF therapy in CD patients using transcriptome imputed from genotypes ²⁷⁹ ; (3) a machine learning to identify, in CD patients, predictive factors of remission and drug durability with ustekinumab ²⁸⁰ ; or (4) a machine learning gene expression to predict response to ustekinumab in CD patients. ²⁸¹

Prediction of side effects

If in precision medicine, biomarkers can be used to predict response to treatment, they can also predict the occurrence of side effects. Some IBD patients (more frequently UC patients) can develop fever and a worsening of diarrhoea under mesalamine. ²⁸² After conducting GWASs followed by a meta-analysis on two independent pharmacogenetic Japanese IBD cohorts (MENDEL and Tohoku), Suzuki et al. identified a significant association between rs144384547 (upstream of RGS17) and these adverse reactions called ‘mesalamine allergies’. ²⁸² In addition, using the GWAS results, they suggest a polygenic risk score and established a combined genetic/clinical prediction model, which yielded a higher area under the curve than polygenic risk score or clinical factors alone (area under the curve, 0.89; sensitivity, 71.4%; specificity, 90.8%). ²⁸² Patients with N-acetyltranferase 2 (NAT2) gene polymorphism (in particular slow metabolizer allele) have a higher risk to develop salazosulfapyridine (a 5-ASA metabolized to sulfapyridine and mesalamine) dose-related adverse effects (NAT2 playing a role in sulfapyridine metabolism).^283,284 The prediction of the risk of life-threatening myelosuppression induced by thiopurine medications using thiopurine methyltransferase (TPMT) measurement is used for several years. ²⁸⁵ The measurement of TPMT activity allows us to adjust the thiopurine dose to avoid this type of side effects and could be superior to genotype to predict the risk of their occurence.^285,286 More recently, Yang et al. found that nudix hydrolase 15 NUDT15 polymorphisms (SNP: rs116855232) were also associated with the onset of myelosuppression induced by thiopurine in a cohort of South Korean patients. ²⁸⁷ A few years later, the achievement of an exome-wide association study in European patients affected and unaffected by thiopurine-induced myelosuppression allowed us to confirm that carriage of any three coding NUDT15 variants was associated with an increased risk of myelosuppression. ²⁸⁸ Similarly, HLA-DQA1-HLA-DRB1 (rs2647087) polymorphism has been associated with the risk of thiopurine-induced pancreatitis^289,290, HLA-DRB1*03:01 with the risk 5-ASA-induced nephrotoxicity ²⁹¹ and HLA-DQA1*05 variant with the risk of development of antibodies to both infliximab and adalimumab.^292,293 Still regarding thiopurine, tobacco and GSTM1-null genotype were risk factors for thiopurines-induced adverse events such as myelosuppression, hepatotoxicity and pancreatitis. ²⁹⁴ Regarding methotrexate, the presence of the methylenetetrahydrofolate reductase (MTHFR) 1298C mutation could be associated with a risk of the occurrence of side effects with methotrexate in IBD patients.^295,296 Another team reported that the presence of HLA-DQ2 risk haplotypes more than doubled the risk of anti-drug antibody (ADA) formation in patients with immune-mediated inflammatory diseases. ²⁹⁷ Finally, Steenholdt et al. ²²² have also shown that the carriage of the minor allele of FASLG, rs76110 increased the risk of severe infusion reactions. The realization of a panel of SNP-based genetic tests before the administration of a treatment could possibly identify patients at risk of side effects and guide the treatment, but would nevertheless require a cost–benefit analysis. ²¹