The Type I Diabetes Genetics Consortium ‘Rapid Response’ family-based candidate gene study: strategy, genes selection, and main outcome

Abstract

Candidate gene studies have long been the principal method for identification of susceptibility genes for type I diabetes (T1D), resulting in the discovery of HLA, INS, PTPN22, CTLA4, and IL2RA. However, many of the initial studies that relied on this strategy were largely underpowered, because of the limitations in genomic information and genotyping technology, as well as the limited size of available cohorts. The Type I Diabetes Genetic Consortium (T1DGC) has established resources to reevaluate earlier reported genes associated with T1D, using its collection of 2298 Caucasian affected sib-pair families (with 11 159 individuals). A total of 382 single-nucleotide polymorphisms (SNPs) located in 21 T1D candidate genes were selected for this study and genotyped in duplicate on two platforms, Illumina and Sequenom. The genes were chosen based on published literature as having been either ‘confirmed’ (replicated) or not (candidates). This study showed several important features of genetic association studies. First, it showed the major impact of small rates of genotyping errors on association statistics. Second, it confirmed associations at INS, PTPN22, IL2RA, IFIH1 (earlier confirmed genes), and CTLA4 (earlier confirmed, with distinct SNPs) loci. Third, it did not find evidence for an association with T1D at SUMO4, despite confirmed association in Asian populations, suggesting the potential for population-specific gene effects. Fourth, at PTPN22, there was evidence for a novel contribution to T1D risk, independent of the replicated effect of the R620W variant. Fifth, among the candidate genes selected for replication, the association of TCF7-P19T with T1D was newly replicated in this study. In summary, this study was able to replicate some genetic effects, reject others, and provide suggestions of association with several of the other candidate genes in stratified analyses (age at onset, HLA status, population of origin). These results have generated additional interesting functional hypotheses that will require further replication in independent cohorts.

Introduction

Until recently, all the genes that were identified to be associated with type I diabetes (T1D) susceptibility originated from candidate gene studies. Only few of these associations were confirmed over all studies performed (for example HLA, INS, PTPN22), whereas others were only replicated in some studies, or not at all. This situation has evolved very recently, as the major technological and methodological progresses occurring in the last few years, and the availability of increasing large cohorts have resulted in the feasibility of genome-wide association scans (GWAS), which now provide the power for hypothesis-free association studies. Although these GWA studies are leading to the identification of an increasing number of T1D-associated single-nucleotide polymorphisms (SNPs), these are often at some distance to the identification of the responsible gene and causal variant, as a consequence of extended linkage disequilibrium regions over the genome. In contrast, candidate genes have been selected based on their known or putative roles in immune function and autoimmunity, or β-cell function, and positive association results provide direct evidence to disease mechanism. Hence, the consideration of candidate genes, with the load of earlier explorations of these genes, appeared as a promising source to identify and/or confirm T1D associated genes or new associated SNPs in known genes, using the best available technology and genomic information, and relying on large cohorts.

Several associations have been reported in candidate genes that have been selected and investigated by many investigators over many years, based on their functional hypotheses regarding T1D mechanisms. Many of these initial association studies were largely underpowered, because of limitations in the size of cohorts or in the extent of SNP coverage in these genes. Hence, a proportion of these positive reports are expected to correspond to false positive results, taking into account the limited statistical support in some cases and the well-known publication bias in the initial reports and replication studies. Nevertheless, and despite these limitations, the combined ‘candidate gene’ status and prior report of association strengthens the candidate status of these selected genes.

For genes whose association with T1D has been reliably replicated across studies, additional questions may be addressed, including the search for additional independent (or not) SNPs that may also contribute to T1D association. In addition, as large-scale genome-wide case/control association studies are in progress and rapidly identifying new associated SNPs and genes, further exploration of these in a large independent family-based cohort is of interest, for replication and for comparison purpose. Replication from an initial non-synonymous SNP scan¹ was also included in this project.

In this study, we performed a comprehensive investigation of earlier reported significant associations of non-MHC genes with T1D, taking advantage of the power of currently available technology, and of the large collection of family material assembled by the Type I Diabetes Genetic Consortium (T1DGC) (Rich et al.,² this volume). The overall aim was to replicate or further explore earlier association reports on candidate genes; in addition, these genotypes and information available in T1DGC families give the possibility to address other questions on the function of these genes in T1D, such as population heterogeneity, the impact of genes depending on subphenotypes, interaction between risk genotypes, and parents of origin effects.

General strategy and study design

Screening population

The study cohort consisted of the T1DGC family panel available at the time of study, comprised initially of 2325 Caucasian affected sib-pair families, which was reduced to 2298 families (11159 DNA samples) after excluding 4 families with unresolved Mendelian inconsistencies and 23 families that were duplicated within the collection. The final family collection included nine independent cohorts, five from established collections from the United Kingdom (British Diabetes Association (BDA)), Denmark, Sardinia, and North America (The Human Biological Data Interchange (HBDI), Joslin), and newly recruited T1DGC families from four regional networks in North America, Europe, United Kingdom, and Asia-Pacific.

Genes and SNPs selection

Valid candidate genes for further exploration were those genes that have been reported to be significantly associated with T1D (P<0.05), excluding genes and SNPs located within the MHC locus, and whose study is the object of a distinct T1DGC study group. On the basis of these criteria, we performed a literature search to select an initial set of genes and SNPs from which we defined two subsets: those with confirmed or replicated status (P<0.05) in at least one independent study (set 1: ‘confirmed’), and those that have not been replicated, either by lack of replication study or failed replication in limited studies (set 2: ‘replication study’).

Different SNP coverage was allocated in these two subgroups of genes, to reflect interest and different screening purposes in these two sets. In the ‘confirmed genes’ study (set 1 genes), the main focus was to search for additional SNPs of interest in addition to further replication of ‘confirmed’ associated SNPs. These additional SNPs included causative SNPs (associated with the earlier reported SNP, possibly with significantly increased association to T1D compared with this SNP) and additional independently associated SNPs. In the replication study (set 2 genes), the focus was primarily to test solely for replication of the specific SNPs that were published. Another question that was considered in this experiment was the possibility of population heterogeneity in T1D susceptibility association, as suggested, for example, by replication of SUMO4 association restricted to date to Asian populations.³ Hence, set 1 genes required ‘full-density coverage’ (with the limit of the currently known SNPs). Thus, selection was based on SNPs with r²<1, including the SNPs that have been reported to be associated with T1D; in contrast, set 2 genes required at least the SNPs that were initially reported to be associated with T1D, and more SNPs to extend the study, as necessary.

From the initial selection of genes and SNPs, a set of 384 SNPs was finalized, with full screening of set 1 genes (r²<1), and partial screening of set 2 genes (as possible, with r² adjusted to complete the 384 SNPs). Finally, additional SNPs reported in a recent non-synonymous SNP scan of T1D¹ were included as the information became available.

Genotyping and quality control

Genotyping was performed at the Broad Institute of MIT and Harvard (http://www.broad.mit.edu), and detailed description of the genotyping procedure, quality control (QC), and data structure is provided in Brown et al.,⁴ this volume. The 384 SNPs were genotyped on two independent platforms, Sequenom (SQNM) and Illumina (ILMN), to achieve very high reliability of genotyping, to reach a very high probability to genotype selected SNPs, particularly those selected for replication study, on at least one platform, and as a pilot experiment for platform evaluation for subsequent genotyping efforts.

Genotyping QC was performed at the Broad Institute, and more detailed QC analyses and basic statistical analyses (genotyping success rate, distortion of heterozygosity, discordance between genotyping platforms, Hardy-Weinberg statistics, transmission disequilibrium test, and pedigree disequilibrium test analyses) were performed at the Coordinating Center at Wake Forest University (Brown et al.,⁴ this volume). Data were released in three consecutive datasets, each completing and revising the earlier release. Data quality was also revised and improved in the consecutive releases. In particular, the latest SQNM release used Typer 4.0 software, providing much increased quality over the earlier released data that used Typer 3.0 software. In addition to 16 SNPs that were used for fingerprinting, as part of QC, a total of 382 SNPs were genotyped, 357 on ILMN, 361 on SQN, and 336 simultaneously on both platforms.

Genetic analyses

The working strategy for analyses of the data adopted by the T1DGC Steering Committee was that gene-specific working groups would conduct a detailed study of the selected genes. An invitation was made by the Coordinating Center to interested researchers to perform detailed gene-specific analyses, including authors who had earlier published on T1D genetic studies of the selected genes (as of September 2006). A general invitation to the T1DGC contributing investigators (who had families included in the Rapid Response project) was made for self-manifested interest. Each working group was charged with leadership in the analysis and interpretation of the function of the specified T1D candidate gene data generated.

Selected genes, specific questions, and data available to working groups

Five genes, INS, PTPN22, CTLA4, IL2R/CD25, and SUMO4, were confirmed for association with T1D susceptibility, based on earlier publications, with an initial association report and at least one significant replication study. Sixteen genes were selected for the purpose of replication study (Table 1).

Table 1.

Selected candidate genes and SNPs genotyped in the T1DGC Rapid Response study

Gene	Chromosome	SNPs initially selected for replication purpose: designation (original code, on cDNA or protein); rs numbera	Number of SNPs genotyped: total (ILMN; SQNM)
Set 1: Full study
INSb	11	−23HphI; rs689*c	21 (18; 17)
		+1140 A/C; rs3842753
PTPN22	1	R620W; rs2476601	28 (26; 25)
CTLA4	2	T17A; rs231775*c	24 (24; 22)
IL2RA/CD25	10	None	69 (66; 65)
SUMO4	6	M55V; rs237025	15 (15; 15)
Set 2: Replication study and partial screening
IL12B	5	3′UTR 1159 C/A; rs3212227	27 (26; 26)
IL4R	16	−3223 G/A; rs2057768	42 (36; 39)
		−1914 T/C; rs2107356
		I50V; rs1805010
		N142N; rs3024571
		E375A; rs1805011
		L389L; rs2234898
		C406R; rs1805012
		S478P; rs1805015
		Q551R; rs1801275
		S761P; rs1805014
IL4 (for IL4R replication study)d	5	−524 C/T; rs2243250	10 (10; 10)
IL13 (for IL4R replication study)d	5	−1512 A/C; rs1881457	5 (3; 5)
		−1112 C/T; rs1800925
		Int3 C/T; rs1295686
		R110Q; rs20541
OAS1	12	splice variant; rs10774671	11 (11; 11)
VDR	12	FokI T/C; rs10735810	40 (38; 39)
		BsmI A/G; rs1544410
		ApaI G/T; rs7975232
		TaqI C/T; rs731236
CXCL12	10	G801A; rs1801157	37 (33; 36)
PAX4	7	P321H; rs712701	10 (10; 10)
FOXP3e	X	None	6 (6; 6)
IRS1	2	G971R;rs1801278	16 (16; 14)
TCF7f	5	P19T; rs5742913	11 (9; 11)
IFIH1	2	A946T; rs1990760	6 (6; 6)
IFIH1 region (intergenic)		rs13422767*
IFIH1 region (intergenic)		rs2111485
IFIH1 region (IFIH1)		H848R; rs3747517
IFIH1 region (GCA)		rs3788964
IFIH1 region (intergenic)		rs984971
IFIH1 region (KCNH7)		rs2068330
CAPSL	5	rs1445898	1 (1; 1)
CEACAM21	19	rs2302188	1 (1; 1)
EFHB	3	rs2929366	1 (1; 1)
Q7Z4C4	5	rs9127	1 (1; 1)

Abbreviations: SNP, single-nucleotide polymorphism; T1DGC, Type I Diabetes Genetics Consortium.

^aSNPs marked with an * were not successfully designed or genotyped in either genotyping platform.

^bThe INS VNTR (−365VNTR) could not be genotyped by these SNP genotyping technologies, but is in almost complete LD with the two selected SNPs in Caucasian populations.

^cINS-23HphI/rs689 and CTLA4-T17A SNPs were genotyped in this collection before the Rapid Response study, as part of the general study protocol.

^dIL4 and IL13 genes were initially selected to allow replication of the interaction study performed for IL4R genes in the original study.

^eThe associated polymorphism based on the published positive report is a (GT)n microsatellite, which could not be genotyped using SNP genotyping technology.

^fThe associated SNP based on the published positive report is referred to as C883A (cDNA), resulting in a P19T non-synonymous amino-acid substitution.

Evaluation of ‘confirmed’ genes

Insulin (INS). A VNTR located 5′ of the insulin gene (INS) was initially reported to be associated to T1D,⁵ later confirmed by numerous studies. Subsequent fine mapping studies showed equally strong association of T1D with two SNPs (−23HphI and + 1140AC) that cannot be significantly discriminated in Caucasian populations.^6,7 The INS VNTR could not be genotyped using the current SNP genotyping technology, so these two SNPs were selected as surrogates. Full-density SNP coverage of INS was performed to explore the hypothesis that VNTR subsets, marked by specific haplotypes, may have variable strength of T1D association as suggested earlier.⁸

Lymphoid tyrosine phosphatase (PTPN22)

Using a candidate-based approach, the PTPN22-R620W non-synonymous SNP was shown to be significantly associated with T1D.⁹ This association was later replicated in numerous studies in Caucasian populations and the same SNP was associated with risk of several other autoimmune diseases. However, the PTPN22-R620W association was not replicated with T1D in Asian and Indian populations,^10,11 in which the 620W variant was absent or found at a low frequency. In addition, the PTPN22 association with T1D was observed with other SNPs in Japanese and Korean populations.¹¹ The PTPN22-R620W SNP was selected for genotyping, as well as full-density SNP coverage to test the hypothesis that additional PTPN22 SNPs may confer independent T1D risk.

Cytotoxic T-lymphocyte antigen 4 (CTLA4)

Many studies have tested SNPs at CTLA4 to evaluate it as a candidate gene for T1D and other autoimmune diseases.¹² However, replication results have been inconsistent across multiple studies.¹³ On the basis of a large study,¹⁴ the association of CTLA4 with T1D was generally confirmed. The combination of the minor risk increase, the small size of initial studies, the different SNPs tested in different studies, and the lack of systematic re-sequencing at this gene may explain, in part, the discordance between studies. In addition to the generally positive association results, there has been suggestive evidence of linkage with T1D to a region containing CTLA4 (IDDM12). These results provided further support for a function of CTLA4 in T1D, although the currently reported T1D-associated SNPs at CTLA4 could not explain the extent of T1D linkage at this locus.¹⁵ Given the status of the confirmation evidence, CTLA4 was included in set 1 genes for examination with full SNP density coverage.

Interleukin-2 receptor (IL2RA/CD25)

Using a systematic multilocus association scan of the candidate gene IL2RA, a significant association with T1D has been identified.¹⁶ The original association of IL2RA with T1D has been replicated in an independent study,¹⁷ with the location of T1D-associated SNPs further refined.¹⁸ Full-density SNP coverage of IL2RA/CD25 was performed in this study.

Small ubiquitin-like modifier 4 (SUMO4)

SUMO4 was tested as a positional candidate gene for the IDDM5 locus on chromosome 6q25. Association of the non-synonymous SNP (M55V) associated with T1D has been published and replicated in Asian populations,^19,20 but not in most studies performed in Caucasian populations. Heterogeneity between populations has been suggested as a rationale for non-replication between populations.²¹ The SUMO4-M55V SNP and full-density SNP coverage of SUMO4 was performed to further test for replication in additional Caucasian populations (using different T1DGC networks and collections), and explore the hypothesis that different SUMO4 SNPs may be associated with T1D in Caucasians.

Genes selected for the ‘replication study’

Positive association results of several other candidate genes have been reported, generally in small-scale studies, which were not tested for replication, or tested but not replicated in independent studies, or with conflicting replication results between studies that overall could not support the gene to the ‘confirmed’ association status. In addition to having been selected as candidate genes, independent observations supporting a functional role of specific disease-associated SNP or variant have been reported for several of these genes, further strengthening the hypothesis of their relevance to T1D susceptibility. These included the following genes (references only to the main positive association studies): IL12B,²² OAS1,²³ VDR,^24,25 CXCL12/SDF1,²⁶ PAX4,²⁷ IRS1,²⁸ TCF7,²⁹ FOXP3.³⁰ In addition, IL4R was tested, together with IL4 and IL13, to replicate earlier studies showing evidence for association of IL4R SNPs and haplotypes, in interaction with IL4 and IL13 SNPs.^31,32

Finally, as the SNPs selection was being finalized for this study, a non-synonymous GWAS for T1D was published,¹ showing strong evidence for association (with internal replication) with the interferon induced with helicase C domain 1 gene (IFIH1), and with several other non-synonymous SNPs located in other genes and unknown transcripts: CAPSL, CEACAM21, EFHB, and Q7Z4C4. The strongest associated SNP at each of these genes was included in the study, and six additional SNPs located in the region of IFIH1, but a complete genetic coverage of all these loci could not be designed at this stage of the project.

The selected genes, selected SNPs, and total number of SNPs genotyped in each gene and region are shown in Table 1. The complete list of SNPs is available through the T1DGC (http://www.T1DBase.org).

QC considerations and impact on association results

Compared with other association studies, this study was performed under stringent QC criteria, using independent genotyping on duplicate platforms that was performed blindly to each other, and using an established QC protocol. Consequently, a very high level of reproducibility was obtained between platforms, with a mean concordance rate of >99% for the 336 markers that were genotyped in duplicate on both platforms. Only three SNPs had genotype concordance rates between platforms <95%.

The association results (based on single marker pedigree disequilibrium tests) were highly concordant, with only one SNP having strongly discordant results (that is P<0.001 and P>0.05 for the same SNP on the two platforms). This SNP, IL4R-rs1805012, was associated with T1D (P = 2 × 10⁻⁸) with ILMN data, but not associated (P>0.05) using SQNM data. This SNP also had a low genotype concordance rate (81%) between platforms. In this case, closer examination of the data showed that the ILMN genotyping was likely to be defective, with rs1805012 minor allele frequency estimated at 0.005 on the ILMN platform, making the ILMN T1D association likely to be spurious.

In general, for SNPs that did not perform well on genotyping, there was clustering of ‘problematic’ features (that is low concordance rates, low success rates, distorted HWE test) and genotypes identified as likely problematic were more likely to show spurious trend for association (P<0.05; data not shown). In the context of duplicate genotyping, the likely problematic platform was easy to identify and the corresponding data excluded from further analyses. Without stringent QC and duplicate samples, these situations become more difficult to detect.

Thus, these observations suggest care in using and interpreting genotype data with any problematic secondary indicator of genotype quality (that is HWE distortion, low genotyping success rate, and low genotype concordance rate, when available), in addition to the standard QC procedure (such as examination of allele clustering), to rule out the possibility of spurious association results. In view of the high quality of genotypes obtained on both platforms, subsequent genotyping were performed on a single platform.

Results

Individual gene studies were performed by the working groups and are presented individually. Briefly, these studies confirm associations with T1D at INS (Howson et al.,³³ this volume), PTPN22 (Steck et al.;³⁴ Howson et al.,³³ this volume), IL2RA (Qu et al.;³⁵ Howson et al.,³³ this volume), and IFIH1 (Howson et al.,³³ this volume). These results confirm those earlier reported in Caucasian populations. The current analyses also support an association at CTLA4 with T1D (Qu et al.;³⁶ Howson et al.,³³ this volume); however, the power of the study remains limited in view of the small odds ratio for the effect at CTLA4. As a consequence, the observed association of CTLA4 with T1D could not be excluded from that with the flanking (candidate) gene, ICOS. Our study did not detect evidence for association at SUMO4 (Podolsky et al.;³⁷ Howson et al.,³³ this volume), despite published replication in Asian populations, suggesting the possibility of a population-specific effect. At PTPN22, there was some support for the function of an independent effect on T1D in addition to R620W association, with an independent haplotype block conferring T1D protection (Steck et al.;³⁴ Howson et al.,³³ this volume).

Among the 10 genes selected for replication study, association was confirmed for the non-synonymous SNP rs5742913 (cDNA C883A, protein P19T) in TCF7, a T-cell transcription factor (Erlich et al.,³⁸ this volume). This result was restricted to the non-DR3/4 subgroup of T1D cases, replicating the primary association report for this SNP with the same stratification criterion. For the remaining genes selected for replication (IL12B, OAS1, VDR, CXCL12/SDF1, PAX4, IRS1, FOXP3, IL4R (taking into account IL4 and IL13 interactions, as performed in the primary report), CAPSL, CEACAM21, EFHB, Q7Z4C4), the initial T1D finding could not be replicated (Morahan et al.;³⁹ Qu et al.;⁴⁰ Kahles et al.;⁴¹ Bergholdt et al.;⁴² Erlich et al.;⁴³ Howson et al.,³³ this volume). Some analyses, however, identified suggestive associations in stratified data (based on age at onset of T1D, HLA status, or population subgroups) (for example Morahan et al.;³⁹ Qu et al.,⁴⁰ this volume). These results may represent interesting features related to disease mechanisms or genetic interactions. At the same time, these exploratory analyses should be taken with caution, considering the number of statistical tests that increases the risk of spurious positive results. These suggestive observations will need to be replicated in independent cohorts.

Discussion

As this study was in progress, GWA studies for T1D have proven efficient and powerful, and have been progressing rapidly in large case–control cohorts. These GWA studies result in the detection of an increasing number of SNPs associated with T1D. These studies also provided replication of association at some of the candidate genes selected in this study, including INS, PTPN22, IL2RA, IFIH1, CTLA4.^1,44,45 Coverage in GWA studies is not complete, as shown by the fact that the replicated TCF7 association has not been detected in GWAS to date. Thus, there remains justification for continued candidate gene study or, alternatively, higher-density SNPs scans in GWAS studies. This strategy would need to be evaluated in terms of cost efficiency and statistical limitations because of the increased number of tests.

As an extension of this study of T1D candidate genes, an additional set of SNPs was selected in genes and regions that have been reported to be associated with T1D (11 regions, including further explorations in the CTLA4/ICOS region), other autoimmune diseases (31 regions), β-cell development genes that were found to be associated with type II diabetes (six genes), the vitamin D-binding receptor GC, and the most significant associated SNPs identified in a recent T1D GWAS (1715 SNPs).⁴⁶ These studies, reported in this volume (Cooper et al.⁴⁷ and Cooper et al.,⁴⁸ this volume) confirmed associations in three known T1D regions (ERBB3 on chromosome 12q13; KIAA1109 on chromosome 4q27; and C12orf30 on chromosome 12q14). In addition, a suggestive association with T1D was also found at three T2D-associated genes (CDKAL1, SLC30A8, and CDKN2B), with SNPs located in CTLA4 (but not its neighboring gene ICOS) and in the vitamin D-binding receptor gene GC. Among the autoimmune selected candidate genes, suggestive association with T1D was also detected at KIF5A, AGAP2, IL7R-UGT3A1 region, SLC22A4, DEFB129, and NCF4 (Cooper et al.,⁴⁷ this volume). These preliminary results will need to be further tested and replicated in larger T1D cohorts.

The GWAS replication study supported association with T1D at four loci that have been identified in other studies (PTPN22 in chromosome 1p13, the chromosome 12q13 region, SH2B3 in chromosome 12q24, CLEC16A in chromosome 16p13, and UBASH3A in chromosome 21q22) (Cooper et al.,⁴⁸ this volume). Three regions were newly identified in the screening stage of this study, and associations at FHOD3 in chromosome 18q12 and at SNP rs5979785 near TLR7 and TLR8 on chromosome Xp22 were further supported when considered jointly with a recent meta-analysis of available T1D GWAS results.

Our study suggests the possibility of population-specific effects. Association at SUMO4 was not replicated in these Caucasian cohorts, and there was no evidence for association of other SUMO4 SNPs (Podolsky et al.³⁷; Howson et al.,³³ this volume). Conversely, PTPN22-R620W association was not confirmed in Asian populations, where this SNP is rare or absent. The test of this hypothesis of population-specific effects will require comparable large-scale studies in well-characterized T1D cohorts. Population-specific effects are generally thought as genetically determined or resulting from gene–environment interactions; in the case of OAS1, the reported association was restricted to the Canadian population (Qu et al.,⁴⁰ this volume). It was proposed that this may result from extended LD blocks in Canadian population compared with other Caucasian populations, extending to a known region of T1D association on chromosome 12q24, also detected in GWAS replication performed in this study. Hence, population history, affecting the genomic architecture, may also result in heterogeneity between populations.

The establishment of the T1DGC, and the resources that it has been able to assemble, allowed performing this comprehensive test of earlier reported gene associations in a common set of subjects. This study provided confirmatory or suggestive results at several genes and loci that had not been identified by other strategies, and also provided evidence supporting gene interaction, specific functional hypotheses, or population events that resulted in some population or subgroup-specific effects. Although candidate genes studies are likely to remain a significant source of new T1D findings, future explorations will undoubtedly need to integrate candidate genes and functional information on genes to systematic genome-wide screening studies of T1D and other relevant diseases and metabolic pathways to increase their power.