Pharmacogenomic and clinical data link non-pharmacokinetic metabolic dysregulation to drug side effect pathogenesis

Drug side effects cause a significant clinical and economic burden. However, mechanisms of drug action underlying side effect pathogenesis remain largely unknown. Here, we integrate pharmacogenomic and clinical data with a human metabolic network and find that non-pharmacokinetic metabolic pathways dysregulated by drugs are linked to the development of side effects. We show such dysregulated metabolic pathways contain genes with sequence variants affecting side effect incidence, play established roles in pathophysiology, have significantly altered activity in corresponding diseases, are susceptible to metabolic inhibitors and are effective targets for therapeutic nutrient supplementation. Our results indicate that metabolic dysregulation represents a common mechanism underlying side effect pathogenesis that is distinct from the role of metabolism in drug clearance. We suggest that elucidating the relationships between the cellular response to drugs, genetic variation of patients and cell metabolism may help managing side effects by personalizing drug prescriptions and nutritional intervention strategies.

glands, and thus the cell culture side effect-linked drug response is recapitulated in vivo in this case 7 . including the phospholipid transporter ABCA1 9 , have been associated with antipsychoticinduced movement disorders including Parkinsonism 10 . Additionally, another study found genes in lipid-associated pathways, among others, to be significantly associated with antipsychoticinduced extrapyramidal symptoms, but these SNPs failed to reach a GWAS significance level 11 . i. Relevant perturbation: choline phosphate (up), prostaglandin E1 (down) ii. Overlaps with gene pathway: We observed two Parkinsonism-linked metabolic gene expression perturbations in lipid pathways, including the phospholipid precursor choline phosphate and a poly-unsaturated fatty acid (PUFA), PGE1. As ZNF202 is known the regulator broad classes of lipids, we included all lipid pathways as potential overlapping pathways, a set of 73/344 metabolites. iii. Relevance of the perturbation to the physiology of the side effect pathology: Multiple physiological ties between altered lipid metabolism and the pathology of Parkinson's disease have been suggested, most notably the link between the lipid binding protein alpha-synuclein, a key component of Lewy bodies 12 , and both altered phospholipids 13 and altered PUFAs 14 . Additionally, the disruption of mitochondrial function associated with perturbation of PUFAs has been proposed to play a role 14 . iv. Perturbation associated with the pathology: Oxidative stress and oxidized lipid products been associated with Parkinson's disease 15 . Substantial lipid alterations are known to occur in the brain during onset of Parkinson's disease 16 . Dietary unsaturated fatty acid intake has been shown to affect risk for Parkinson's disease 17 . v. Known supplements targeted at perturbed pathways: Multiple studies have shown essential fatty acids to be a beneficial supplement in the treatment of Parkinson's disease 18,19 . vi. In vivo occurrence of metabolic perturbation: Essential fatty acid levels have been shown to be altered in red blood cells in vivo under antipsychotic treatment 20 , showing that the Parkinsonism-linked perturbation seen in cell culture has been shown to occur in vivo as well.  21 , have been associated with susceptibility to tardive dyskinesia as an adverse effect of antipsychotic treatment i. Relevant perturbation: formyl-N-acetyl-5-methoxykynurenamine (down) ii. Overlaps with gene pathway: We observe a down-regulation of formyl-N-acetyl-5methoxykynurenamine, the product of a five reaction biosynthetic pathway from tryptophan that includes serotonin. From this pathway, a set of 6/344 metabolites is obtained, as with the MC4R case above. iii. Relevance of the perturbation to the physiology of the side effect pathology: While dopaminergic neuron activity thought to play a central role in the development of tardive dyskinesia, the pathogenesis is not fully elucidated. Serotonin receptors are thought to interact with dopaminergic neurons in tardive dyskinesia 22 , and down-regulation of 5-HT production has been found to be sufficient, though not necessary, for the alleviation of tardive dyskinesia that occurs with deep brain stimulation therapy 23 . iv. Perturbation associated with the pathology: Tardive dyskinesia is not normally described as a pathology independent of drug treatment. That said, serotonin is thought to play a role in tardive dyskinesia 22 , and agonists of the serotonin receptor 5-HT1A have been shown to alleviate tardive dyskinesia, suggesting the serotonin pathway is involved in the pathology or at least affects it 24 . v. Known supplements targeted at perturbed pathways: Results are conflicting as to whether increased or decreased serotonin activity is beneficial, as mentioned above, with decreased 5-HT activity 23 , 5-HT3 antagonism 25 and 5-HT1A agonism 24 all being associated with effective treatment of tardive dyskinesia. A small five patient study testing the efficacy of the serotonin precursor 5-hydroxytryptophan showed no improvement in tardive dyskinesia 26 , which may have been expected given several studies showing decreased 5-HT activity associated with alleviated tardive dyskinesia. vi. In vivo occurrence of metabolic perturbation: Levels of the serotonin metabolite 5-hydroxyindoleacetate has been shown to be perturbed in brain tissues of rats with tardive dyskinesia induced by treatment with the antipsychotic haloperidol 27 . a. Gene: Variants of the D3 dopamine receptor have been shown to be associated with tardive dyskinesia 28 . One other study did find no association between certain dopamine receptor variants and tardive dyskinesia occurrence 29 . i. Relevant perturbation: oxalate (up) ii. Overlaps with gene pathway: Dopamine is synthesized in a sequential set of reactions downstream of tyrosine and converted to norepinephrine by dopamine betamonooxygenase, an L-ascorbate (vitamin C)-dependent enzyme. Including the entire subnetwork as well as the cofactors vitamin C, tetrahydrobiopterin, and S-adenosyl-Lmethionine metabolic subnetworks, we obtain a set of 16/344 overlapping metabolites. iii. Relevance of the perturbation to the physiology of the side effect pathology: Dopamine receptors are thought to play a primary role in the pathogenesis of antipsychotic-induced tardive dyskinesia, as many antipsychotics are dopamine antagonists. We observe an upregulation of the vitamin C metabolite oxalate, indicating an up-regulation of vitamin C metabolism and therefore a possible depletion of L-ascorbate. As L-ascorbate is a cofactor required in the metabolism of dopamine, it is possible that perturbations in vitamin C processing could impact dopamine metabolism. Supporting this, multiple studies have shown that vitamin C has an antidopaminergic effect in combination with antipsychotics 30, 31 , consistent with greater L-ascorbate enhancing dopamine degradation. Thus, it is possible that antipsychotic-induced depletion of vitamin C may alter the drug effect on dopamine receptors, leading to the side effect. iv. Perturbation associated with the pathology: We did not find any links between vitamin C deficiency and tardive dyskinesia independent of drug treatment, as the pathology of the side effect is not reported independently of drug treatment. v. Known supplements targeted at perturbed pathways: Vitamin C supplementation has been shown in combination with Vitamin E to be an effective supplement for tardive dyskinesia, although this effect is thought to be primarily due to antioxidative effects of the vitamins 32 . Furthermore, vitamin C supplementation has been found to improve outcome and decrease stress associated with atypical antipsychotics 33 . vii. In vivo occurrence of metabolic perturbation: Vitamin C has been found to be diminished in schizophrenics, and it is proposed the schizophrenics require greater vitamin C intake than healthy patients 34, 35 . This is consistent with the up-regulated vitamin C degradation pathway seen in drugs that induce tardive dyskinesia, a list dominated by antipsychotics. Furthermore, levels of dopamine metabolites have been shown to be perturbed in brain tissues of rats with tardive dyskinesia induced by treatment with the antipsychotic haloperidol 27 . Zinc finger homeobox 3 (ZFHX3) have been associated with drug-induced torsades de pointes 36 . Additionally, GPD1L mutations cause Brugada syndrome type 2, a disorder characterized by cardiac arrhythmia 37 , and ZFHX3 mutation has been associated with atrial fibrillation 38 . We group these genes together due to their overlapping role in regulation of oxidative stress response, as described below. i. Relevant perturbation: D-Ribulose 5-phosphate (down) ii. Overlaps with gene pathway: GLPD1 has been shown to be a metabolic oxidative state sensor in the heart 39 . Similarly, ZFHX3 is an oxidative stress response regulator 40 whose mutation is associated with cardiac arrhythmia. Ribulose-5-phosphate is the end product of the oxidative branch of the pentose phosphate pathway that is responsible for a large fraction of NADPH production and maintenance of the reduced glutathione pool and hence the oxidative state of the cell. Thus, we observe down-regulated oxidative pentose phosphate pathway in drugs causing arrhythmia, which overlaps with the known susceptibility genes with functions in sensing cellular oxidation state. We attempted to include all pathways related to oxidative stress, including pentose phosphate pathway, the primary generator of glutathione, hydrogen peroxide, and antioxidant pathways including vitamin C, glutathione, melatonin, lipoate, and coenzyme Q10, a set of 30/344 metabolites. iii. Relevance of the perturbation to the physiology of the side effect pathology: Oxidative state, through NADH state, has been shown to influence the PKC-dependent phosphorylation of the sodium channel SCN5A in GPD1L-linked arrhythmias 39 . Similarly, oxidative stress has been shown to alter K+ currents in the heart 41 in a glutathione-dependent manner. Additionally, oxidative stress is thought to affect mitochondrial energetics in the heart, contributing to arrhythmias 42-44 . iv. Perturbation associated with the pathology: Hydrogen peroxide has been shown to induced cardiac arrhythmias in guinea pig hearts 45 , suggesting oxidative stress can induce arrhythmias directly. v. Known supplements targeted at perturbed pathways:

Drug-induced arrhythmia
Vitamin A and Vitamin C both have been shown to reduce epinephrine-induced arrhythmias 46 . Several antioxidants have been shown effective in reducing reperfusioninduced arrhythmias, including the glutathione precursor N-acetylcysteine 47 , taurine 48,49 , and melatonin 50 . Additionally, the antioxidant coenzyme Q10 has been shown to improve outcome following congestive heart failure 51 . vi. In vivo occurrence of metabolic perturbation: Clomipramine, an antidepressant with one of the highest reported incidences of induced arrhythmia according to the SIDER database, has been shown to induce cardiotoxicity in an oxidative stress-dependent manner 52 , and thus the arrhythmia-linked perturbation in cell culture appears relevant to in vivo drug response in at least one case.  55 . We observe two up-regulations around the urea cycle and arginine metabolism (L-glutamate 5-phosphate and urea), which connects nitric oxide, glutamate, creatine, urea, and polyamine metabolism. As NOS1AP is a nitrogen regulator, we included metabolites involved in the urea cycle, including creatine and nitric oxide, downstream of glutamate and until the polyamine precursor L-ornithine and the first polyamine putrescine, a set of 12/344 metabolites. iii. Relevance of the perturbation to the physiology of the side effect pathology: Neuronical nitric oxide synthase (nNOS), which is bound by NOS1AP, regulates cardiac function in various ways 55 . NO is reported to affect cardiac ion channels and signaling pathways in a variety of ways 56 . iv. Perturbation associated with the pathology: Decreases in cardiac function in cirrhosis have been shown to be NO dependent 57 . Inversely, cirrhotic rats have been found to have increased resistance to epinephrine-induced arrhythmia due to constitutively higher NO production 58 . v. Known supplements targeted at perturbed pathways: Supplementation with L-arginine has been shown to diminish arrhythmia scores in digoxin-treated isolated heart experiments 59 . i. In vivo occurrence of metabolic perturbation: Theophylline, the drug with the highest reported incidence of arrhythmia, has been shown to inhibit arginine-dependent production of NO 60 . Similarly, serotonin blockers such as clomipramine (which, as mentioned above, is associated with a high incidence of arrhythmia) have been shown to inhibit nitric oxide synthase 61 . These findings are consistent with altered nitrogen metabolism observed in cell culture response. e. Links between pathology and other observed metabolic perturbations: As an interesting note, in addition to pentose phosphate pathway and nitrogen metabolism, we also observed linked between the other observed arrhythmia-linked metabolic perturbations in eicosanoid metabolism and glycolysation and the pathogenesis of cardiac arrhythmias. For example, we observe a downregulation of the glycosylation precursor UDP-D-xylose, and altered ion channel glycosylation has been shown to be one mechanism important to drug-induced arrhythmia 62 . Furthermore, we observe an up-regulation of prostaglandin E 1 , which has been shown to be a therapeutic supplement to arrhythmias 63,64 . Additionally, prostaglandins have been shown to affect cardiac electrophysiology 65 , and disturbed prostaglandin ratios post-infarction are proposed to play a role in occurrence of arrhythmias 66 . pathway and forms of coenzyme Q10 are considered to be associated with COQ2, however, from this pathway only ubiquinol-10 itself appears as a potential prediction of the method, resulting in only 1/344 metabolites overlapping with the susceptibility gene. iii. Relevance of the perturbation to the physiology of the side effect pathology:

Statin-induced myotoxicity
Mitochondrial dysfunction is thought to play a role in statin-induced myopathies, and as coenzyme Q10 plays a key role in proper mitochondrial function, induction of Q10 deficiency could clearly play a causal role in disrupting mitochondrial function 68,69 . iv. Perturbation associated with the pathology: Certain coenzyme Q10 deficiencies are known to be associated with myopathies 70 . Mitochondrial myopathy has been found to be associated with coenzyme Q10 deficiency as well 71 , and Q10 supplementation has been found to alleviate this form of myopathy 72 . v. Known supplements targeted at perturbed pathways: Interestingly, the use of Q10 has been shown effective in relieving statin-induced myopathies in some studies 73,74 and ineffective in others 75 . We note that specifically Q10 has been shown to alleviate muscle inflammation (myositis) following exercise independent of statin treatment, suggesting the possibility that the beneficial effects of Q10 are mediated through anti-inflammatory effects 76 , while other pathological factors may be in play in other myopathies. vi. In vivo occurrence of metabolic perturbation: Levels of coenzyme Q10 have been shown in various studies to be decreased in patients undergoing statin treatment [77][78][79][80][81] . i. Relevant perturbation: L-threonate (down) ii. Overlaps with gene pathway: L-threonate is a metabolite of L-ascorbate. COMT is involved in the metabolism of catecholamines, including dopamine, epinephrine, and norepinephrine. Dopamine is synthesized in a sequential set of reactions downstream of tyrosine and converted to norepinephrine by dopamine beta-monooxygenase, an Lascorbate (vitamin C)-dependent enzyme. Including the entire subnetwork as well as the cofactors vitamin C, tetrahydrobiopterin, and S-adenosyl-L-methionine metabolic subnetworks, we obtain a set of 16/344 overlapping metabolites. The vitamin C pathway, where we observed a perturbation, thus overlaps with COMT through its role as a cofactor for dopamine beta hydroxylase. iii. Relevance of the perturbation to the physiology of the side effect pathology: There is significant dopaminergic neuron innervation in the inner ear 83 , and thus proper dopamine metabolism is essential for function. D2/D3 receptors have been found to play an important role in hearing, with dopamine agonists playing a protective role in hearing loss 84 . The production of norepinephrine from dopamine is catalyzed by dopamine beta hydroxylase, which requires vitamin C as a substrate. iv. Perturbation associated with the pathology: Inhibition of vitamin C transport has been reported to cause inhibition of catecholamine production 85 , which interferes with proper dopaminergic transmission. Additionally, mice deficient in dopamine beta hydroxylase, an enzyme in dopamine metabolism that uses vitamin C as a cofactor, have been reported to be subject to acquired hearing loss. v. Known supplements targeted at perturbed pathways: Vitamin C has been shown to be an effective supplement in preventing damage-induced hearing loss 86 . Vitamin C has even specifically been shown to inhibit cisplatin toxicity 87 . In addition, dopamine has been shown to have a protective effect in hearing loss and is thought to be a potentially effective therapeutic supplement 88 . vi. In vivo occurrence of metabolic perturbation: Cisplatin has been shown in vivo in mammals to inhibit secretion of catecholamines 89 , although inhibition of vitamin C metabolism as a possible mechanism for this has not been directly investigated. d. Gene: Variants of glutathione-S-transferases 90  iv. Perturbation associated with the pathology: A deficiency in the ROS response protein superoxide dismutase has been associated with increased susceptibility to hearing loss 92 . v. Known supplements targeted at perturbed pathways: Vitamin C 86 and lipoate 93 have been shown to be effective in preventing noise-and drug-induced hearing loss, respectively, when supplemented. vi. In vivo occurrence of metabolic perturbation: Cisplatin has been observed in vivo to cause a significant oxidative load 94 , which is thought to be a causal factor in its ototoxicity.

Calculation of joint probability of predictions overlapping with side effect susceptibility genes
As a number of perturbations were observed for each side effect, and several metabolites would be considered potentially overlapping in each case (other than with coenzyme Q10, which had a single overlapping metabolite within the set of observed side effect-linked metabolites), it is difficult to determine by inspection whether the side effect-linked metabolites are significantly predictive of genetic susceptibility pathways for corresponding side effects. To attempt to address this rigorously, we took the set of distinct side effect-linked metabolites observed in any case, a set of 344 metabolites, and the set of distinct metabolites linked to each particular side effect, and considered the latter 'draws' in a hypergeometric test. As we observed overlap in 9/9 cases, the joint probability of this occurring by chance is thus the product of the probabilities in each separate case, assessed by hypergeometric tests based on the number of metabolite perturbations and potentially overlapping metabolites in each case. This assumes that the probability of 'drawing' each metabolite is the same as drawing any other, which depends on complex factors determining the underlying prediction distribution; however, we do not expect this to greatly affect the probability, given that apparent distribution of side effect-linked metabolites is relatively uniform upon inspection.
This process resulted in the following statistical tests (p = 1 -cdf + pdf/2): shown to be a susceptibility gene for myotoxicity 95 .

1) Antipsychotic-induced weight gain with
i. Relevant perturbation: prostaglandin E2 (up), prostaglandin F2alpha (down) ii. Overlaps with gene pathway: While SLCO1B1 has been shown to have a role in statin transport, other drugs similarly transport statins, and we note an overlap with gene expression changes in the pathways related to native function of SLCO1B1. We included this study to expose potential overlap between pharmacokinetic and pharmacodynamics effects of gene polymorphisms and drug-induced expression changes. iii. Relevance of the perturbation to the physiology of the side effect pathology: Abnormal lipid oxidation state has been shown to be a predisposing factor for drug-induced myotoxicity 96 . iv. Perturbation associated with the pathology: Higher levels of prostaglandin E2 were found in horses with halothane-induced myotoxicity compared with those not experiencing myotoxicity 97 , suggesting this perturbation is tied to the side effect occurrence. v. Known supplements targeted at perturbed pathways: Supplementation with essential fatty acids has been shown to alleviate asthenia in mice 98 , further suggesting that there may be an overlap between drug metabolism and native metabolic function as perturbed by statins. vi. In vivo occurrence of metabolic perturbation: In vivo eicosanoid metabolism has been shown to be perturbed in muscle following simvastatin and atorvastatin treatment 99 , suggesting that the metabolic gene expression perturbation is physiologically relevant in the primary tissue of interest in myotoxicity.
b. Other related genes: CYP2C8, a cytochrome P450 enzyme responsible for statin degradation, has been shown to be a susceptibility gene for rhabdomyolysis 100 . Its native function is thought to be responsible for arachidonic acid metabolism 101 , and thus its function also overlaps with the observed perturbations in eicosanoid metabolism. shown to result in clotting deficiency in neonates 105 . v. Known supplements targeted at perturbed pathways: Intake of n-6 and n-3 fatty acids has been shown to have either a proaggregatory or an antiaggregatory effect, respectively 106 . vi. In vivo occurrence of metabolic perturbation: We did not find existing evidence of warfarin decreasing in vivo eicosanoid levels.

Side effect not included in study
Rationale: An insufficient number of gene expression samples (< 30) from the Connectivity Map database were from drugs with a particular side effect, and so we were unable to generate predictions for the side effect.

Metabolic pathway not included in study
Rationale: Certain metabolic pathways were 'off' during the MetChange algorithm analysis, due to the assumptions of cell function and media constraints, and hence no predictions were made in these pathways. Notably, this includes heme synthesis and degradation.

Genes without known or overlapping metabolic function
Rationale: In cases where the gene has unknown or non-specific links to metabolism, we were unable to associate the gene with a particular pathway, and no comparison could be made. This includes pharmacokinetic genes that had no known overlapping native function, such as thiopurine methyltransferase.

Side effects with pathology manifested in enucleated cells
Rationale: Genes with polymorphisms were linked to red blood cell anemias in certain drugs. However, as the red blood cell is enucleated and thus would experience no direct drug-induced gene expression changes, we excluded these cases from comparison with side effect -linked gene expression changes. We do not exclude the possibility that nucleated progenitor cells may experience drug-induced expression changes relevant to the pathology (see ribavirin-induced anemia), but it is more difficult to make meaningful hypotheses about such effects, and therefore we ignore them in these cases.
Gene polymorphism likely to impact side effect occurrence through altered drug pharmacokinetics only rather than altered metabolic gene expression Rationale: Studies involving CYPs, drug metabolizing enzymes 107 , and other drug metabolism and transport proteins were generally excluded in cases of non-specific toxicity. First, as the genes are pharmacokinetic and polymorphism affects susceptibility to multiple side effects, we deem it unlikely that the gene is related to the pathology of all such side effects, but rather the effect is manifested primarily through altered pharmacokinetics. Second, while the pharmacokinetic genes are associated with side effect susceptibility of particular drugs, our predictions for a given side effect include gene expression samples from multiple drugs, many of which are likely not metabolized by the same gene, and thus we are not likely to see any conserved gene expression change related to the particular pharmacokinetic gene associated with the drug in the genetic study.

Excluded Studies: Side effect not included in study
1) Drug-induced liver injury a. Gene: Various 108 i. Reason for exclusion: There were not enough gene expression samples (< 30) from drugs that induce liver injury to obtain a consensus signature, so no predictions were made. 2) Pamidronate and zoledronate-induced osteonecrosis of the jaw a. Gene: CYP2C8 is associated with susceptibility to drug-induced osteonecrosis of the jaw 109 i. Reason for exclusion: There were not enough gene expression samples (< 30) from drugs that induce osteonecrosis of the jaw to obtain a consensus signature, so no predictions were made. Additionally, the CYP2C8 gene is tied to metabolism of the drugs, and thus since other drugs not metabolized by CYP2C8 also cause osteonecrosis of the jaw are included, the side effect would not be expected to have a consensus signature linked to this gene.

Excluded Studies: Metabolic pathway not included in study
1) Irinotecan-induced neutropenia a. Gene: UGT1A1 has been shown to be associated with irinotecan-induced neutropenia 110 i. Reason for exclusion: Native function of UGT1A1 is thought to be primarily in bilirubin glucurosylation. We did not include the bilirubin pathway in our analysis and thus had no potential predictions for comparison. Furthermore, UGT1A1 is thought to mediate susceptibility through its role in irinotecan metabolism, so gene expression changes related to UGT1A1-related pathways are not likely to be critical to side effect incidence. 2) Methemoglobin reductase linked to methemoglobinemia and porphobilinogen deaminase deficiency linked to acute porphyric crisis 111 were similarly excluded because the hemoglobin synthesis and degradation pathway was not included in our analysis 3) Statin-induced myotoxicity a. Gene: ATP-binding cassette G2 112 i. Reason for exclusion: Altered drug transport is thought to be the primary mechanism of increased susceptibility with gene polymorphisms, and thus no overlap with induced gene expression changes is expected. Native function appears to be related to porphyrin transport 113 , which is a pathway that we had excluded from analysis and thus could not examine possible overlap.
Excluded Studies: Genes without known or overlapping metabolic function 1) Succinylcholine-induced apnea a. Gene: Serum pseudocholinesterase has been lined to apnea induced by succinylcholine 114 i. Reason for exclusion: The native substrate for pseudocholinesterase is not in our model, and thus there was no overlapping pathway with which to associate the gene. Additionally, the mechanism of sensitivity is thought to be through the role of the cholinesterase in metabolizing succinylcholine, and thus we did not expect to see a causal role of gene expression changes related to choline pathways in the side effect pathogenesis.
2) 6-mercaptopurine-induced hematological toxicity and cisplatin-induced ototoxicity a. Gene: TPMT, encoding thiopurine methyltransferase, is associated with toxicity of the chemotherapeutic agent 6-mercaptopurine 115 . Also, variants of TPMT have been associated with cisplatin-induced hearing loss 82 . i. Reason for exclusion: The effect of TPMT polymorphisms on 6-mercaptopurine toxicity is likely primarily related to 6-mercaptopurine metabolism, but the enzyme has no native substrate identified to our knowledge 116 , and thus we do not have an overlapping pathway. 3) Citalopram-induced blurry vision a. Gene: Emid2 is associated with citalopram-induced visual and hearing defects, including blurry vision 117 i. Reason for exclusion: Function of Emid2 is not fully elucidated and no clear ties to metabolism could be found for comparison 118 4) Nevirapine-induced rash b. Gene: CCHCR1 has been shown to be associated with nevirapine-induced rash 119 i. Reason for exclusion: No clear tie between CCHCR1 and a particular metabolic pathway could be found. 5) Drug-induced cardiac arrhythmia c. Gene: CERKL (ceramide kinase-like protein) is associated with iloperidone-induced QT prolongation 120 i. Reason for exclusion: CERKL has an unclear function and thus we were unable to determine whether the gene has ties to metabolic function d. Gene: HERG, a potassium channel, has been shown to be associated with drug-induced torsades de pointes 121 i. Reason for exclusion: As a potassium channel, links between HERG and metabolism are non-specific and thus the case was excluded. e. Gene: KvLQT1, a potassium channel, has been shown to be associated with long QT syndrome 122 i. Reason for exclusion: As a potassium channel, links between KvLQT1 and metabolism are non-specific and thus the case was excluded. f. Gene: SCN5A, a calcium channel, has been shown to be associated with long QT syndrome 123 i. Reason for exclusion: As a calcium channel, links between SCN5A and metabolism are non-specific and thus the case was excluded. 6) Anesthetic and muscle relaxant-induced malignant hyperthermia g. Gene: RYR1, a ryanodine receptor functioning as a calcium channel, has been associated with drug-induced malignant hypothermia 124 i. Reason for exclusion: Links between RYR1 and metabolism are unclear and could not be tied to a particular pathway 7) Contraceptive-induced venous thrombosis h. Gene: Factor V, a coagulation factor, has been associated with venous thrombosis in patients taking oral contraceptives 125 i. Reason for exclusion: Coagulation is outside of the scope of the metabolism and so no clear overlap could be established. 8) Doxorubicin-induced cardiotoxicity i. Gene: Top2b, a topoisomerase, has been shown be protective from doxorubicin-induced heart failure 126 i. Reason for exclusion: Links between Top2b and metabolism are not direct, and thus we could not associate the gene with a particular metabolic pathway. 9) Antipsychotic-induced weight gain j. Gene: Various genes, including MEIS2, PRKAR2B, GPR98, FHOD3, RNF144A, ASTN2, SOX5, and ATF7IP2, were found to be associated with metabolic side effects of antipsychotics in a genome-wide association study 127 i. Reason for exclusion: While MEIS2 and PRKAR2B are reported as having ties to metabolism, no specific metabolic pathway could be identified as related to these genes. 10) Epirubicin-induced leukopenia/neutropenia k. Gene: MCPH1, a gene thought to be associated with DNA damage response 128 , was associated with epirubicin-induced leukopenia 129 i. Reason for exclusion: Beyond a generic role as a DNA damage response and cell cycle checkpoint protein, links between MCPH1 and metabolism are unknown. 11) Abacavir-induced hypersensitivity a. Gene: HLA-B*5701 was identified as a risk factor for abacavir-induced hypersensitivity 130 i. Reason for exclusion: Links between metabolism and immune hypersensitivity are nonspecific and could not be tied to a particular pathway 12) For reviews on other studies correlated drug toxicity with immune gene polymorphisms: see reviews [131][132][133] .
Excluded Studies: Pathology manifested in enucleated cell 1) Drug-induced anemia a. Gene: Variants of ITPA have been shown to protect against ribavirin-induced anemia 134 via alterations of ITP levels in the red blood cell 135 i. Reason for exclusion: The pathology of hemolytic anemia is manifested at the level of the red blood cell, which is enucleated and thus is not expected to have a gene expression response relevant to drug action. That said, we did notice an increase of ITPA expression that was greatest in the HL-60 hematopoietic line, which, if relevant to the side effect pathology, would be consistent with the protective effect of ITPA deficiency. Thus, we note the possibility that ITPA expression could be increased in RBC progenitor cells as a response to ribavirin treatment, dysregulating nucleotide levels and increasing susceptibility to hemolytic anemia. This is merely conjecture, however, given that the level of ITPA activity has not been compared in ribavirin-treated and control RBC samples, to our knowledge. b. Gene: GAPD deficiency is classically associated with drug-induced anemia 136 i. Reason for exclusion: Pathology manifested primarily through oxidative stress generated from a pentose phosphate pathway deficiency 136 Excluded Studies: Gene polymorphism likely to impact side effect occurrence through altered drug pharmacokinetics only rather than altered metabolic gene expression 1) Methotrexate-induced toxicities a. Gene: MTHFR, methylenetetrahydrofolate reductase, is associated with various methotrexateinduced toxicities 137 i. Reason for exclusion: Toxicity includes multiple pathologies, and thus could not be tied to a particular side effect. It is assumed therefore that the effect of polymorphisms of MTHFR on side effect incident is primarily related to pharmacokinetic effect, and therefore side effect-linked gene expression changes would not be expected to overlap with the native pathway of MTHFR 2) Fluorouracil-induced toxicity a. Gene: Dihydropyrimidine dehydrogenase, DPYD, deficiency has been shown to be associated with 5-fluorouracil toxicity 138 i. Reason for exclusion: Toxicity associated with fluorouracil is non-specific and occurs through a variety of pathologies 139 , and thus we could not tie it to a particular side effect. For this reason, it is assumed that DYPD affects primarily the pharmacokinetics of the drug, and thus we do not expect a functional overlap between DYPD and gene expression changes associated with a particular side effect. 3) N-acetyltransferase-linked toxicities a. Gene: N-acetyltransferases are linked to various toxicities in various drugs 140 i. Reason for exclusion: Due to the fact that multiple side effects are associated with polymorphisms in the same gene, we assume that the effect of polymorphisms on susceptibility is primarily through their effect on metabolism of the drugs.