Surrogate Genetics and Metabolic Profiling for Characterization of Human Disease Alleles

Abstract

Cystathionine-β-synthase (CBS) deficiency is a human genetic disease causing homocystinuria, thrombosis, mental retardation, and a suite of other devastating manifestations. Early detection coupled with dietary modification greatly reduces pathology, but the response to treatment differs with the allele of CBS. A better understanding of the relationship between allelic variants and protein function will improve both diagnosis and treatment. To this end, we tested the function of 84 CBS alleles previously sequenced from patients with homocystinuria by ortholog replacement in Saccharomyces cerevisiae. Within this clinically associated set, 15% of variant alleles were indistinguishable from the predominant CBS allele in function, suggesting enzymatic activity was retained. An additional 37% of the alleles were partially functional or could be rescued by cofactor supplementation in the growth medium. This large class included alleles rescued by elevated levels of the cofactor vitamin B6, but also alleles rescued by elevated heme, a second CBS cofactor. Measurement of the metabolite levels in CBS-substituted yeast grown with different B6 levels using LC–MS revealed changes in metabolism that propagated beyond the substrate and product of CBS. Production of the critical antioxidant glutathione through the CBS pathway was greatly decreased when CBS function was restricted through genetic, cofactor, or substrate restriction, a metabolic consequence with implications for treatment.

THE first complete human genome sequence seeded the defining challenge of human genetics for the foreseeable future: interpreting the impact of variations in the sequences of individual human genomes. Comparative genome sequencing reveals an average of one single-nucleotide change per 1200 bp between any two individuals. In the absence of strong Mendelian inheritance and linkage, confirming that any human genotype actually caused a phenotype is a significant challenge given the approximately 3 million genetic variants per person. Indeed, 4000 traits of medical interest show evidence for inheritance but lack a clear determinant (Online Mendelian Inheritance in Man 2012). Next-generation sequencing within small pedigrees (Ng et al. 2010a,b; Fan et al. 2011), or a more narrowly defined clinical phenotype (Schubert et al. 1997), can sometimes disentangle the underlying contribution of a gene to disease. In this work we have taken an approach that complements both increased sequencing capacity and expanded phenotypic description. We used surrogate genetics to assay directly the function of allelic variants and then evaluate their potential contribution to phenotypes of clinical importance.

Homocystinuria, elevated levels of the sulfur-containing metabolite homocystine in the urine, illustrates several challenges inherent to elucidating the molecular bases of human genetic diseases. Worldwide, 1 in 335,000 individuals are affected (Mudd et al. 1995), but the frequency approaches 1 in 1800 in certain populations (Gan-Schreier et al. 2010). A few well-characterized alleles of the gene encoding cystathionine β-synthase (CBS) correlate with disease symptoms, providing an appealing molecular mechanism. The enzyme CBS converts homocysteine to cystathionine in the cysteine biosynthesis pathway (Supporting Information, Figure S1). In people with homocystinuria, free homocysteine accumulates and can covalently bind to proteins or oxidize to the dimer homocystine. Disease indicators include homocystinuria or hyperhomocysteinemia, an abnormally high concentration of serum total homocysteine, the sum of free, oxidized, and protein-bound forms.

CBS catalyzes a committed step in the pathway that produces cysteine and ultimately glutathione, the major endogenous intracellular antioxidant. Upstream of CBS, homocysteine is an intermediate in the pathway that recycles S-adenosylmethionine (AdoMet), the major methyl donor in the cell, back to methionine. The wide range of symptoms may reflect the fact that CBS and its variants have the potential to alter regulatory methylation of DNA and histones, as well as the redox state of the cell. Yet, elevated homocysteine levels occur in many people, including heterozygotes for some CBS alleles, without any clinical symptoms (Motulsky 1996; Guttormsen et al. 2001). Additionally, defects in several different genes tangential to cysteine biosynthesis, such as MTHFR, can lead to homocysteinemia and similar symptoms (Frosst et al. 1995; Gaughan et al. 2001; Pare et al. 2009). Hence, elevated homocysteine level is a convenient marker for a metabolic imbalance, but the cause and consequences may be elusive.

The genetic contributions are complex, but because early medical intervention, including a diet low in protein and methionine, successfully alleviates many homocystinuria symptoms, neonatal screening is widespread (Mudd et al. 2001). Vitamin supplementation can replace dietary restriction as a therapy in a highly allele-dependent manner. CBS uses a vitamin B6 cofactor to form cystathionine by the condensation of serine and homocysteine. Hence, elevated B6 is thought to partially compensate for vitamin-responsive alleles with a lower affinity for the B6 cofactor (Chen et al. 2006). Human CBS also forms multimers, coordinates heme with a bound iron, and contains a regulatory domain that binds the metabolite AdoMet as a possible regulatory mechanism (Shan and Kruger 1998; Meier et al. 2001; Christopher et al. 2002; Scott et al. 2004; Chen et al. 2006; Sen and Banerjee 2007). These features suggest control points for enzyme regulation and function, or targets for nutritional and pharmaceutical therapies, that CBS alleles may affect differently.

Directed sequencing efforts of patients afflicted with homocystinuria have produced a large catalog of alleles (Kraus et al. 2012), with both common and rare alleles (Mudd et al. 1985; Kraus 1994; Gallagher et al. 1995, 1998). However, clinical association does not guarantee causality. In many cases, the sequenced alleles are further analyzed by genetic or biochemical means, providing most of our knowledge of CBS deficiency. Despite these heroic efforts, the piecemeal identification of alleles, variations in assessment strategies, diploid nature of the human genome, and increasing numbers of rare alleles all lead to uncharacterized alleles that may cause subtle, but important, differences in phenotype. As ever more CBS alleles are found, the need for reliable measures of allele impact will increase. CYS4 is the Saccharomyces cerevisiae ortholog of CBS and has the same function in yeast as in humans (Ono et al. 1988). Although yeast Cys4p lacks a heme binding domain and may differ in details of its biochemical regulation, human CBS complements cys4 yeast for cysteine and glutathione production (Kruger and Cox 1994, 1995). Furthermore, nonfunctional or B6-remedial CBS alleles recapitulate their human phenotypes in yeast cys4 mutants (Kim et al. 1997; Shan and Kruger 1998). We took advantage of the foundation built by previous, elegant cross-species complementation experiments (Kruger and Cox 1994, 1995) to develop a quantitative, comprehensive, and direct test of how variation in a single human disease gene correlated with disease and treatment via nutritional supplementation.

Materials and Methods

Plasmids

The plasmid pHUCBS was the kind gift of Warren Kruger and served as the template for generating alternative CBS alleles using the QuikChange II Kit (Agilent). We selected single-base pair missense mutations from the CBS Mutation Database (Kraus et al. 1999, 2012), from published literature, and from the RefSeq database for A69P (rs17849313), P70L (rs2229413), and R369P (rs11700812). We verified the sequence of the entire open reading frame of each allele (Table 1). The pHUCBS plasmid and all subsequent clones contain a single, silent base pair change (909C > T) relative to the RefSeq sequence for CBS (L14577.1). A BstEII/FseI fragment containing CBS variants was subcloned between the S. cerevisiae TEF1 promoter and CYC1 terminator in pJR2983, a CEN–ARS URA3 shuttle plasmid.

Table 1. CBS alleles tested for function in yeast.

pJR	Mutation
	Protein	cDNA	Initial citation and clinical characterization, or RefSeq number
pJR3044	H65R	194A > G	Janosik et al. (2001b)
pJR3045	A69P	205G > C	rs17849313
pJR3046	P70L	209C > T	rs2229413
pJR3047	P78R	233C > G	de Franchis et al. (1994)
pJR3048	G85R	253G > C	Maclean et al. (2002)
pJR3049	T87N	260C > A	Kraus et al. (2012)
pJR3050	P88S	262C > T	Sebastio et al. (1995)
pJR3051	L101P	302T > C	Gallagher et al. (1998)
pJR3052	K102Q	304A > C	Kozich et al. (1997)
pJR3053	K102N	306G > C	de Franchis et al. (1994)
pJR3054	C109R	325T > C	Gaustadnes et al. (2002)
pJR3055	A114V	341C > T	Kozich et al. (1993)
pJR3056	G116R	346G > A	Sperandeo et al. (1996)
pJR3057	R121C	361C > T	Katsushima et al. (2006)
pJR3058	R121H	362G > A	Bermudez et al. (2006); Katsushima et al. (2006)
pJR3059	R121L	362G > T	Kraus et al. (2012)
pJR3060	M126V	376A > G	de Franchis et al. (1999)
pJR3061	E128D	384G > T	Coude et al. (1998)
pJR3062	E131D	393G > C	Marble et al. (1994)
pJR3063	G139R	415G > A	Shih et al. (1995)
pJR3064	I143M	429C > G	Orendae et al. (2004)
pJR3065	E144K	430G > A	Shih et al. (1995)
pJR3066	P145L	434C > T	Kozich et al. (1993)
pJR3067	G148R	442G > C	Orendae et al. (2004)
pJR3068	G151R	451G > A	Kraus et al. (2012)
pJR3069	I152M	456C > G	Kraus et al. (2012)
pJR3070	G153R	457G > C	Kraus et al. (2012)
pJR3071	L154Q	461T > A	Lee et al. (2005)
pJR3072	A155T	463G > A	Janosik et al. (2001b)
pJR3073	A155V	464C > T	Lee et al. (2005)
pJR3074	A158V	473C > T	Shan and Kruger (1998)
pJR3075	C165Y	494G > A	Kluijtmans et al. (1995)
pJR3076	V168M	502G > A	Kraus et al. (2012)
pJR3077	E176K	526G > A	Kozich et al. (1997)
pJR3078	V180A	539T > C	Kluijtmans et al. (1999)
pJR3079	T191M	572C > T	Urreizti et al. (2003)
pJR3080	D198V	593A > T	Kraus et al. (2012)
pJR3081	R224H	671G > A	Kruger and Cox (1995)
pJR3082	A226T	676G > A	Kruger et al. (2003)
pJR3083	N228S	683A > G	Kruger et al. (2003)
pJR3084	N228K	684C > A	Gallagher et al. (1998)
pJR3085	D234N	700G > A	De Lucca and Casique (2004)
pJR3086	E239K	715G > A	de Franchis et al. (1994)
pJR3087	T257M	770C > T	Sebastio et al. (1995)
pJR3088	G259S	775G > A	Kraus et al. (2012)
pJR3089	T262M	785C > T	Kim et al. (1997)
pJR3090	R266G	796A > G	Katsushima et al. (2006)
pJR3091	R266K	797G > A	Kim et al. (1997)
pJR3092	C275Y	824G > A	Urreizti et al. (2003)
pJR3093	I278T	833T > C	Kozich and Kraus (1992)
pJR3094	A288T	862G > A	Lee et al. (2005)
pJR3095	A288P	862G > C	Linnebank et al. (2004)
pJR3096	P290L	869C > T	De Lucca and Casique (2004)
pJR3097	E302K	904G > A	Sperandeo et al. (1996)
pJR3098	G307S	919G > A	Hu et al. (1993)
pJR3099	V320A	959T > C	Kim et al. (1997)
pJR3100	A331E	992C > A	Dawson et al. (1997)
pJR3101	A331V	992C > T	Kruger and Cox (1995)
pJR3102	R336H	1007G > A	Coude et al. (1998)
pJR3103	G347S	1039G > A	Gaustadnes et al. (2002)
pJR3104	S349N	1046G > A	Urreizti et al. (2003)
pJR3105	S352N	1055G > A	Dawson et al. (1997)
pJR3106	T353M	1058C > T	Dawson et al. (1997)
pJR3107	V354M	1060G > A	Coude et al. (1998)
pJR3108	A355P	1063G > C	Gallagher et al. (1998)
pJR3109	A361T	1081G > A	Castro et al. (1999)
pJR3110	R369C	1105C > T	Kim et al. (1997)
pJR3111	R369H	1106G > A	Kraus et al. (2012)
pJR3112	R369P	1106G > C	rs11700812
pJR3113	C370Y	1109G > A	Tsai et al. (1997)
pJR3114	V371M	1111G > A	Kluijtmans et al. (1999)
pJR3115	D376N	1126G > A	Kruger et al. (2003)
pJR3116	R379W	1135C > T	Linnebank et al. (2004)
pJR3117	K384E	1150A > G	Aral et al. (1997)
pJR3118	K384N	1152G > T	Kraus et al. (2012)
pJR3119	M391I	1173G > A	Kraus et al. (2012)
pJR3120	P422L	1265C > T	Maclean et al. (2002)
pJR3121	T434N	1301C > A	Kraus et al. (2012)
pJR3122	I435T	1304T > C	Maclean et al. (2002)
pJR3123	R439Q	1316G > A	Dawson et al. (1997); Tsai et al. (1997)
pJR3124	D444N	1330G > A	Kluijtmans et al. (1996)
pJR3125	L456P	1367T > C	Urreizti et al. (2003)
pJR3126	S466L	1397C > T	Janosik et al. (2001a)
pJR3127	Q526K	1572C > A	Kruger et al. (2003)

Strains

All S. cerevisiae strains serving as a host for a human CBS allele contained a complete deletion of CYS4 (MATα cys4Δ::KanMX his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0, JRY9292) derived from the yeast knockout collection (Winzeler et al. 1999). A hem1cys4 strain was created by disruption of HEM1 with LEU2 (hem1Δ::LEU2). CBS transformants were selected by uracil prototrophy.

Growth assays

Strains containing CBS plasmids were maintained on complete synthetic medium lacking uracil (CSM-Ura) and supplemented with glutathione, a stable source of cysteine. cys4 complementation was assayed by growth on solid CSM-Ura medium without glutathione and with 400 ng/ml vitamin B6 (pyridoxine–HCl). CBS alleles that complemented cys4 were further characterized in a quantitative growth assay using a minimal liquid medium made with yeast nitrogen base lacking vitamin B6 or other vitamins and amino acids (MP Biomedicals). Vitamins (biotin, pantothenate, inositol, niacin, p-aminobenzoic acid, riboflavin, and thiamin) were included at standard concentrations, and vitamin B6 was supplemented at six different concentrations: 0, 0.5, 1, 2, 4, and 400 ng/ml B6. Histidine, leucine, and lysine were added to relieve auxotrophies in the parent strain, and methionine was included in all minimal media. Growth rate assays used 250 μl volumes and started with cells at OD₆₀₀ = 0.002, inoculated from cells pregrown in minimal medium that lacked B6 and contained glutathione. The pregrowth medium in the hem1 experiments contained 50 μg/ml δ-aminolevulinc acid (δ-ALA) and the cells were washed twice with minimal medium to prevent carry-over. For the sake of clarity, we refer to supplementation with the soluble heme precursor δ-ALA as “heme supplementation” in the text. Heme is sparingly soluble and δ-ALA supplementation was more efficient. The optical density (A₆₀₀ nm) was measured every 30 min for 96 hr at 28° in a stationary microplate reader (Molecular Devices VersaMax). We accounted for settling of the cells over time by resuspending the cells after the final kinetic read and measuring the OD₆₀₀ value. Data were then normalized using the time-weighted ratio of the endpoint kinetic OD₆₀₀ value to the resuspended OD₆₀₀ value, according to the formula,

$$\left(\frac{\text{Endpoint}\hspace{0.17em}\text{kinetic}\hspace{0.17em}\text{OD}}{\text{Resuspended}\hspace{0.17em}\text{OD}}\hspace{0.17em}-1\right)\frac{\text{time}\hspace{0.17em}\text{point}}{\text{final}\hspace{0.17em}\text{time}\hspace{0.17em}\text{point}}+1,$$

and were log₁₀ transformed. Due to the stationary-phase ceiling, growth rate better described the growth of alleles than end-point measurement. Growth rate was calculated as the slope of the regression line for data values between OD₆₀₀ = 0.05 and OD₆₀₀ = 0.1. Data were compared to the major allele grown in the same medium on the same plate. Outliers were detected using Grubb’s test and removed from growth rate calculations. The raw growth rate data are available as File S1.

Immunoblots for CBS protein quantification

The total protein concentration of boiled, NaOH-extracted yeast pellets was measured using the Pierce BCA Protein Assay Kit (Thermo Scientific) to normalize sample concentrations. Proteins were visualized on an Odyssey Infrared Imager (Li-COR Bioscience) after separation on a denaturing gel. Mouse anti-CBS polyclonal antibody (Abnova H00000875-A01), a rabbit anti-3-phosphoglycerate kinase (PGK) antibody (a gift from Jeremy Thorner, University of California Berkeley) and an anti-proliferating cell nuclear antigen (PCNA) antibody (Abcam ab70472) were used to detect target proteins.

Metabolite measurements

Cells were cultured in liquid minimal medium that contained glutathione before washing with, and inoculation into, minimal medium lacking glutathione. Equal numbers of cells from log-phase cultures were harvested 12 hr after inoculation. Metabolite extraction combined previously described methods (Canelas et al. 2008; Boer et al. 2010; Godat et al. 2010) as follows: 8.0 × 10⁸ (G307S data set) or 1.9 × 10⁹ (V320A data set) cells were pelleted by centrifugation at 3200 × g. The cell pellets were resuspended with 9.5 ml of their spent medium supernatant, then quenched with 20 ml −80° methanol. The cells were pelleted at 4000 × g at −10° in a rotor (Sorvall SS-34), prechilled to −80°, and then resuspended with 1.0 ml of 4° extraction solvent [0.1% perchloric acid with 400 μM glycine-1-¹³C,¹⁵N (Sigma 299340) and 20 μM isotopically labeled methionine-¹³C₅,¹⁵N (Sigma 608106)]. The samples were boiled for 5 min, and cell debris and precipitated proteins were removed by centrifugation for 2 min at 4000 × g in a 4° microfuge. The supernatants were diluted 1:4 in 0.1% perchloric acid and 0.1% formic acid. Liquid chromatography–mass spectrometry (LC–MS) analysis used 20-μl injection volumes. Chromatographic separation (2.1 × 250 mm, 5 μm Discovery HS-F5 column; Supelco) used a water-to-acetonitrile gradient (Godat et al. 2010) and was followed by detection on an LTQ-Orbitrap XL hybrid mass spectrometer equipped with an IonMax electrospray ionization source (Thermo Fisher Scientific, Waltham, MA). For the G307S data set, a fourfold dilution series of a mixture of 17 metabolite standards was added to a pooled cell extract that contained equal volumes from each experimental sample, and was then used for metabolite identification and calibration. A full calibration panel was included in the V320A experiment, but was not added to a pooled standard. LC–MS data were converted to centroids and the mzXML file format using ReAdW 4.3.1 (Deutsch et al. 2010) with an Xcalibur library (ThermoFisher Scientifics, v. 2.0.7). Peak processing used the BioConductor package XCMS (Smith et al. 2006; Tautenhahn et al. 2008); processed data are available as File S2. Metabolites were identified using the pooled calibration standards and the Human Metabolome Database (Wishart et al. 2009) for the G307S study and by exact mass only for the V320A analysis. Zeros in the data were imputed using local minima, data were normalized using upper quartiles, and intensities were log transformed for analysis using R (scripts and centroided data files are provided as File S3).

Results

The surrogate assessment of clinically associated CBS alleles in Saccharomyces cerevisiae

We selected all alleles of CBS documented prior to 2011 that could be generated by a single base-pair change and that affected an amino acid (Table 1). Each human CBS allele was synthesized, inserted into a yeast plasmid, and individually transformed into a cys4 yeast strain lacking the CBS ortholog CYS4. Centromere-based vectors were used to reduce copy-number variation. Eighty-one alleles derived from patients with homocystinuria plus three additional variants found in public databases were assayed. This collection of 84 missense mutations included alterations in the heme-binding, catalytic, and AdoMet-binding regulatory domains of the CBS protein. This strain could not grow on minimal media, but the defect in cysteine biosynthesis was bypassed by the addition of cysteine or the more stable downstream metabolite glutathione. Critically, the endogenous CYS4 gene supports more robust growth than any CBS allele, suggesting that CBS function was rate-limiting in yeast. Therefore, all assays necessarily compared CBS alleles to the major allele of human CBS (major allele, MA), not to CYS4 (Figure 1 and Figure S2).

Figure 1.

Growth of CBS-complemented yeast on solid media. Cultures grown to saturation in liquid minimal medium containing glutathione and lacking B6 were plated in a fivefold dilution series onto solid medium ± glutathione. Growth was imaged after 3 days at 30°. The growth of the major allele and representative alleles of the nonfunctional and B6-responsive classes are shown.

We discriminated functional from nonfunctional CBS alleles by plating cells onto media containing or lacking glutathione: only CBS alleles that restored CYS4 function supported growth on either medium. Of the 84 alleles, 46 required glutathione supplementation to support growth, indicating severe loss of function (listed as “nonfunctional” in Table 2). Disease alleles often encode misfolded proteins (Yue et al. 2005), and there is precedence for lower protein levels among some nonfunctional CBS alleles due to aggregation (Katsushima et al. 2006) or degradation (de Franchis et al. 1994; Urreizti et al. 2006; Singh et al. 2007, 2010). Nonetheless, while we inferred misfolding or aggregation of some CBS proteins, degradation may differ in yeast and human cells, perhaps because an appropriate E3 ligase is missing. We observed ample steady-state levels of the CBS protein encoded by 17/17 different human CBS alleles, representative of different growth classes, regardless of B6 availability, as determined by immunoblot (Figure 2, Figure S3, and Table 2). Hence, our data measured the effect of mutations on the intrinsic functions of the enzyme without complication from protein turnover.

Table 2. Summary of clinical and yeast phenotypes.

Mutation	Clinical Response to B6	Enzyme activity	Yeast phenotype and remediation
H65R	Nonvariable	Lowa	Nonfunctionalo
A69P	NA		Similar to major allele
P70L	NA		Intermediate growth, B6 remedial, hem1 rescue
P78R	Variable	Meda,b	Similar to major allele
G85R	Partial response	Lowc	High growth, B6 and heme remedial
T87N	ND		Low growtho
P88S	ND		Low growtho
L101P	Conflicting	Lowd	Nonfunctional
K102N	Variable	Meda,b	Similar to major alleleo
K102Q	ND		Similar to major allele
C109R	Conflicting	Lowd	Nonfunctionalo
A114V	Conflicting	Mede/higha	Low growth
G116R	Variable		Nonfunctional
R121C	ND		Nonfunctional
R121L	ND		Nonfunctional
R121H	Nonvariable		Nonfunctional
M126V	Nonvariable		Nonfunctionalo
E128D	Nonvariable		Similar to major allele
E131D	Nonvariable	Lowf	Low growtho
G139R	Variable		Intermediate growtho
I143M	Nonvariable	Lowg	Nonfunctional
E144K	Conflicting	Lowa,h	Nonfunctionalo
P145L	Nonvariable	Lowi	Nonfunctionalo
G148R	Nonvariable	Lowa	Nonfunctional
G151R	ND		Nonfunctionalo
I152M	Conflicting	Lowj	hem1 rescue
G153R	NA		Nonfunctional
L154Q	NA	Lowk	Nonfunctional
A155T	Conflicting		Low growth, B6 and heme remedial
A155V	NA	Lowk	hem1 rescue
A158V	NA		Low growth
C165Y	Conflicting	Lowa,j	Nonfunctional
V168M	ND		Nonfunctionalo
E176K	Nonvariable	Lowa	hem1 rescue
V180A	Variable	Meda,j	Intermediate growth, B6 remedial
T191M	Nonvariable	Lowa,j	Nonfunctional
D198V	Nonvariable		Low growtho
R224H	ND		Low growth
A226T	Variable	Medl	Low growtho
N228S	Nonvariable		Nonfunctional
N228K	Nonvariable	Lowa,c	Nonfunctional
D234N	Nonvariable		Intermediate growth, heme remedial
E239K	Variable		Nonfunctional
T257M	Nonvariable		Nonfunctional
G259S	ND		Nonfunctional
T262M	Nonvariable		Low growth, B6 remedial
R266G	Nonvariable		Nonfunctional
R266K	Variable	Meda	High growth, B6 and heme remedial
C275Y	Nonvariable	Lowl	Nonfunctional
I278T	Conflicting	Lowa,m	Low growth
A288P	NA		Nonfunctional
A288T	NA	Lowk	Nonfunctional
P290L	Variable		hem1 rescue
E302K	Conflicting	Lowe/higha	Nonfunctional
G307S	Conflicting	Lowa,n	Nonfunctionalo
V320A	Conflicting		Intermediate growth, B6 and heme remedial
A331E	ND	Lowh	Nonfunctional
A331V	ND		Low growth
R336H	Variable	Lowl	hem1 rescue; not tested in liquid media
G347S	Variable	Lowd	Nonfunctional
S349N	Nonvariable	Lowl	Nonfunctional
S352N	Nonvariable		Low growth
T353M	Conflicting	Lowh	Low growth, B6 and heme remedial
V354M	Nonvariable		High growth, B6 remedial
A355P	ND		Nonfunctional
A361T	Nonresponsive		Nonfunctional
R369P	NA		Nonfunctional
R369H	ND		Similar to major allele
R369C	Responsive	Lowj/meda	Similar to major allele
C370Y	Responsive		Nonfunctional
V371M	Partial response	Lowj	Similar to major allele
D376N	ND		Nonfunctional
R379W	ND		hem1 rescue
K384E	Responsive		Nonfunctionalo
K384N	Nonresponsive		hem1 rescue
M391I	Nonresponsive		Nonfunctional
P422L	B6 nonresponsive	Higha,c	Similar to major allele
T434N	B6 responsive		High growth, B6 remedial
I435T	ND	Higha,c	Similar to major allele
R439Q	Conflicting	Meda	Similar to major allele
D444N	Conflicting	Higha/medl	Similar to major allele
L456P	B6 nonresponsive	Lowl	Intermediate growth
S466L	ND	Higha,c	Similar to major allele
Q526K	B6 nonresponsive		Intermediate growth, hem1 rescueo

Alleles with yeast growth phenotypes inconsistent with clinical (underlined) or in vitro activity (italics) are indicated. Alleles identified in the clinic but without information about B6 response (ND, no data) and alleles identified from available sequence only (NA, not applicable) are indicated. Enzyme activity summarizes several different assays including expression in Escherichia coli or in human fibroblast cell culture. Low indicates activity at or below the level of detection, med indicates an intermediate activity and high indicates activity indistinguishable from the major allele.

^aKozich et al. (2010).

^bde Franchis et al. (1994).

^cMaclean et al. (2002).

^dGaustadnes et al. (2002).

^ede Franchis et al. (1999).

^fMarble et al. (1994).

^gOrendae et al. (2004).

^hDawson et al. (1997).

ⁱKozich et al. (1993).

^jKraus et al. (1999).

^kLee et al. (2005).

^lUrreizti et al. (2006).

^mKozich and Kraus (1992).

ⁿHu et al. (1993).

^oThe alleles we analyzed by immunoblot. The original publication and a recent reanalysis were cited.

Figure 2.

CBS protein levels in yeast whole cell extracts. (A) Immunoblotting of yeast cells with the CBS major allele (MA), a B6-responsive allele (A226T), an AdoMet-domain mutation (Q526K), or an empty expression vector (EV) were grown in minimal medium with 400 ng/ml B6 alone, with glutathione alone, or with glutathione and 400 ng/ml B6. (B) Yeast cells with the CBS MA and five variant alleles were grown in minimal medium with glutathione alone or with glutathione and 400 ng/ml B6. Representative alleles from the nonfunctional (T87N and P88S) and sick (P145L, V168M, and M126V) phenotypic classes were processed for immunoblotting. 3-Phosphoglycerate kinase (PGK) was detected as a loading control.

Although many of the alleles tested were identified in individuals with clinically significant homocysteinemia, 38 CBS alleles were capable of supporting growth on medium lacking glutathione and hence retained substantial function (alleles that are not listed as nonfunctional in Table 2). These alleles were further assayed in liquid medium at varying concentrations of vitamin B6 to expand the qualitative phenotype to a quantitative assessment of function and B6 responsiveness. All CBS alleles grew poorly without B6 supplementation (Table S1). Although S. cerevisiae is a B6 prototroph, the endogenous B6 level was insufficient to support the B6 requirement of human CBS. However, cells with the major CBS allele grew relatively well in medium supplemented with as little as 1 ng/ml of B6 (Figure S2 and Figure 3A).

Figure 3.

CBS yeast exhibited B6-dependent growth. (A) Representative growth curves of yeast with the major allele of human CBS cultures supplemented with six different levels of B6 (colored lines). Average growth rate (±SD) is shown for each B6 level (n = 84–90). (B) The growth of each mutant (n ≥ 4) was expressed as the percentage of average growth rate of yeast with the major allele of human CBS at each B6 level (±SD).

When compared to cells with the major allele, the growth phenotypes of cells with other CBS alleles varied greatly (Figure 3B and Table S1). Hierarchical clustering by growth rate under all conditions was used to describe allele behavior. This nonbiased method separated alleles into roughly three bins, in addition to the nonfunctional bin defined above (Figure 4A). Cells with 14 different CBS alleles had evidence of some function but grew poorly even at high levels (400 ng/ml) of B6 (listed as “low growth” in Table 2). Cells with 7 alleles showed an intermediate phenotype with growth rates between that of cells with the major allele and cells with poorly functioning CBS (listed as “intermediate growth” in Table 2B). Cells with each of the remaining 17 alleles grew at rates similar to those of cells with the predominant allele (listed as “high growth” or “similar to major allele” in Table 2). Ten alleles, spanning all growth classes, shifted from a lower growth-rate class to a higher growth-rate class at 400 ng/ml B6 (listed as “B6 remedial” in Table 2). All functional alleles benefitted from increased B6 concentrations; however, cells with these 10 alleles were especially sensitive.

Figure 4.

CBS yeast growth responses to B6 and heme grouped alleles into distinct classes. Heat maps of growth rates normalized to the growth of the major allele after titration of (A) B6 in HEM1 yeast or (B) B6 and heme in hem1 yeast. The column Z-score indicates the mean growth rate (Z-score of 0) and standard deviation (Z-score of ±1) of all alleles per column, with positive Z-scores indicating higher than average growth. Arrowheads indicate alleles that respond to cofactor titration more strongly than other alleles in their cluster. Asterisks (*) denote alleles that failed to grow in HEM1 yeast but were capable of growth in hem1 yeast.

The importance of cofactor concentration to CBS function extended to heme

The CBS enzyme coordinates a second cofactor, heme, through a heme-binding domain. Certain mutations in the heme-binding domain disrupt CBS function in human cells, indicating that heme is critical to protein activity (Janosik et al. 2001b). Furthermore, heme increases the activity and dynamics of some CBS alleles (Kopecka et al. 2011). Indeed, one of the mutations in our set, H65R, alters a heme-coordinating residue and was not functional in yeast. In contrast, S. cerevisiae Cys4p lacks a heme-binding domain and does not require heme. Yeast produce heme for other purposes, and the media in our previous experiments lacked additional heme. Therefore, endogenous heme production was sufficient for human CBS function. We hypothesized that some alleles of CBS might be heme responsive under sufficiently challenging conditions. We tested this hypothesis using a yeast hem1 strain that was unable to synthesize δ-aminolevulinic acid (δ-ALA), the first committed step in heme biosynthesis, and was therefore a heme auxotroph. We varied in vivo heme levels by amending the medium with δ-ALA and determined the affect on CBS function.

A two-cofactor titration of all alleles in the hem1 background revealed intriguing information about the heme cofactor and about allele function (Figure 4B and Table S1 and Figure S4). hem1 yeast with the predominant CBS allele were incapable of growth without heme supplementation and showed growth dose dependence on both B6 and heme levels. Likewise, strains with each of the 38 alleles with measurable growth in the B6-only titration grew better in media with higher heme concentrations, suggesting heme was required for CBS function and was limiting for growth in hem1 yeast. Cells with 6 alleles grew worse than the predominant allele at 2.5 ng/ml heme, but had growth rates approaching that of cells with the predominant CBS allele at 50 ng/ml heme, indicating that some defective CBS variants were especially sensitive to heme (listed as “heme remedial” in Table 2). Five of these heme-responsive alleles were also remedial with B6, apparently identifying proteins whose deficiency could benefit from increased concentration of either cofactor. The D234N allele alone benefited more from increased heme than from increased B6.

The remaining 32 alleles were no more sensitive to heme than the predominant CBS allele, with two interesting exceptions. Cells with the P70L and Q526K alleles clustered with the low-growth alleles in the HEM1 background but with the predominant CBS allele in the hem1 background. Similarly, cells with 7 of the 46 alleles that appeared nonfunctional in the HEM1 strain grew in medium containing high B6 and high heme, albeit poorly, revealing partial function of these alleles (Figure 4B; listed as “hem1 rescue” in Table 2). Although counterintuitive, rescue of allele function in the hem1 strain may occur because the hem1 mutation induced heme uptake (Protchenko et al. 2008) or increased substrate availability. The dynamic range of CBS-dependent growth in the hem1 background was larger than in a HEM1 strain, manifested as both saturation at higher cell density and better growth at lower B6 concentration (Figure S4).

CBS alleles with clinical association but no apparent defect

The majority of alleles tested supported less growth than the major allele, as might be expected for disease-causing alleles, yet 13 appeared indistinguishable from the major allele (listed as “similar to major allele” in Table 2). Since yeast are typically grown at 30°, we considered the possibility that these 13 alleles encoded temperature-sensitive mutant proteins whose defects were not apparent at lower temperature. We tested the growth of 10 nominally benign substitutions and found that none had growth defects at 37° (Figure S5), nor on medium containing the denaturant formamide, which can reveal partial loss of function (Aguilera 1994). One allele, A69P, even appeared less sensitive to denaturing stress than the predominant allele. Therefore, these alleles encoded fully functional enzymes within the limits of this assay.

Intracellular metabolic imbalances caused by CBS variants

Our previous assays for CBS function relied on growth as a proxy for enzymatic function. As an independent assessment of CBS function, we used LC–MS to measure directly the metabolite profiles of cells with different CBS alleles grown in medium with different levels of B6. Metabolite levels mirrored the trends observed in the growth data: cells with the major CBS allele grown under B6 limitation and cells with a nonfunctional CBS allele induced similar metabolic profiles that differed from profiles of cells with the major CBS allele grown under nonlimiting B6 conditions (Figure 5, Table S1, and File S2). Yeast cells carrying the G307S allele, a nonfunctional allele in clinical and yeast growth assays, failed to produce glutathione and instead accumulated homocystine, sharing the namesake diagnostic phenotype of homocystinuria.

Figure 5.

Metabolite profiles of CBS yeast grown under nutrient replete or limiting conditions. Heat map of amino acid or derivative metabolite levels in cell extracts from yeast grown with either the major CBS allele (MA) or the G307S (nonfunctional) allele, as measured by mass spectrometry. Each column represents the average of four biological replicates. B6 was supplemented at doses that produced robust growth of the major allele (400 ng/ml) or measurable, but compromised, growth (1 ng/ml). Metabolite levels were scaled for each row and both metabolites and experimental conditions were subject to hierarchical clustering. The row Z-score indicates the mean and standard deviations for each metabolite, such that the mean metabolite level has as a Z-score of 0. Duplicate columns were independent cell extracts and demonstrated trial-to-trial variation that was not significant in any of the known metabolites (t-test P > 0.05). The oxidizing conditions used for extraction strongly favored isolation of homocystine over homocysteine. Similarly cystathionine and cysteine were not detected because of limitations in sample processing or because intracellular pools are small.

Analysis of a second allele class, represented by the B6 remedial V320A allele, further defined the correlation between growth rate and metabolite flux (Table S2). Cells relying on the V320A allele accumulated significantly more homocystine and produced less glutathione than the major allele regardless of B6 level. However, in contrast to the nonfunctional G307S allele, glutathione production increased and homocystine accumulation decreased when cells with the V320A allele were grown with a high dose of B6. These data revealed perfect concordance with the relative growth rates of these alleles at these doses of B6 (Figure 3 and Table S1).

The massively parallel nature of LC–MS allowed us to measure metabolites upstream and downstream of CBS, as well as those in shunt pathways. The accumulation of upstream metabolites was not restricted to homocystine in cells with the G307S allele or under B6 limitation of cells with the major CBS allele. Elevated levels of AdoMet, SAH, and methionine, the substrates involved in homocysteine recycling by one-carbon metabolism, were detected (Figure 5 and Figure 6, A–D). Overall, our data suggested that the metabolic footprint of CBS deficiency extended far beyond the immediate substrate and product of the enzyme, homocysteine, and cystathionine, respectively. For example, the block at CBS, through mutation or B6 limitation, caused a detectable drop in 5′-methylthioadenosine, a metabolite in the methionine salvage pathway with the potential to reduce homocysteine levels in favor of increased methionine. Instead, flux through this pathway was also reduced (Figure 5).

Figure 6.

Levels of metabolites critical in CBS function. Scatter plots of the levels of four different metabolites measured by mass spectrometry. The average of four biological replicates (bars) and their individual measurements (squares) are shown. Duplicated columns show trial-to-trial variation in independent cell extracts. (A) Methionine, (B) AdoMet, (C) homocystine, and (D) glutathione. The levels of all four metabolites are significantly different (ANOVA P < 0.0001); all significant differences between the MA at high B6 and other classes are indicated (Tukey’s honest significance test **, P < 0.005; ****, P < 0.0001).

The growth rate of cells with functional CBS alleles in B6-supplemented medium was significantly greater than that in medium lacking B6 or in cells with a loss-of-function allele. To distinguish the metabolic signature of loss of CBS function from the signature of lack of growth per se, we profiled cells with the major allele under methionine starvation. Although the csy4 yeast strain used in these assays synthesizes methionine, additional methionine supplementation was necessary for growth of all CBS-substituted strains. The growth defect without exogenous methionine is as severe as without B6; however, the metabolic profile was strikingly different. Specifically, cells limited for methionine produced low levels of glutathione, but without homocystine accumulation, regardless of B6 concentration (Figure 5 and Figure 6, C and D). These data confirmed that CBS-deficiency generated a unique metabolic profile not due simply to poor growth.

Discussion

Building on Garrod’s Inborn Errors of Metabolism (Garrod 1909), technological innovations have shaped our understanding of how an individual’s genetics cause disease. The Human Genome Project facilitated rapid progress in linking genes and diseases, but also exposed a gap between an increasing number of minor associations and an actual assessment of causality (Bansal et al. 2010; Cirulli and Goldstein 2010; McClellan and King 2010). The so-called “missing heritability” lies, in part, in the failure to define disease with sufficient phenotypic precision. Here, we developed techniques that provided a quantitative assessment of clinically associated alleles that confirmed some expectations and led to unexpected insights about one human genetic disease and presumptive causative alleles.

Using yeast growth, we quantified the relative function of 84 alleles of human CBS, binning alleles according to growth rate and ability to be rescued by B6 or heme cofactors. We also measured the levels of metabolites in cells with three different human CBS alleles by LC–MS, confirming that yeast growth was a relevant proxy for enzyme function and revealing the tight coupling between trans-sulfuration pathway flux and growth. These quantitative phenotypes confirmed that many clinically associated CBS alleles are indeed nonfunctional, with a few notable exceptions. Although computational prediction may eventually replace or supplement laboratory research in the corroboration of genetic associations, the exceptions derived from functional studies offer a starting point for future analyses of protein function and disease (Wei et al. 2010). Similar primary culture-independent, quantitative assays for human alleles in a surrogate organism should be broadly applicable to any gene that fits into an orthologous pathway (Shan et al. 1999; Zhang et al. 2003; Marini et al. 2008). Methods like this are increasingly important given the expanding sequence landscape: since 2010, 38 novel missense alleles of CBS have been identified (NHLBI Exome Sequencing Project 2012).

Eighty-one of the alleles we tested were identified in people with homocystinuria. For 33 alleles, there either are no clinical data about B6-responsiveness or the evidence is conflicting: our data could help to resolve some of these cases. For example, the K102Q allele functioned similarly to the major allele in our growth assay. Recent exome sequencing revealed that this allele, rare in previously sequenced populations, has an allele frequency close to 4% in the African-American population (NHLBI Exome Sequencing Project 2012). Therefore, additional information about this allele is critical to assessing disease risk. For the remaining alleles, growth in yeast and clinical data correlated well, especially for alleles identified in patients who were B6 nonresponsive (Table 2). In addition to clinical data, the in vitro enzymatic activities of many CBS alleles have been assessed. Our growth-rate measurements were consistent with published biochemical studies in 36 of the 40 cases of overlap (exceptions are italicized in Table 2).

Sixteen of the 81 alleles had clinical features that did not match our yeast growth data (underlined in Table 2). Some discrepancies may have resulted from an unrecognized second mutation in CBS in the patient. Additionally, rare alleles generally occurred in a single individual, heterozygous with a different allele, making it difficult to assess the individual connection to disease. However, there may be interesting cases in which CBS function in yeast and humans differ. For example, the P422L and S466L mutations in the C-terminal regulatory domain encode biochemically active proteins that are unable to bind AdoMet and cause a distinctive, mild form of homocystinuria (Maclean et al. 2002). We tested both of these alleles plus six other AdoMet domain mutations and found that all supported growth, suggesting that AdoMet regulation may not be critical for growth in yeast. However, cells with the L456P and Q526K alleles, both altering the AdoMet domain, had reduced growth, while cells with the T434N allele were B6 responsive, indicating that some mutations in the AdoMet domain diminish CBS function. AdoMet regulatory mutations accounted for some, but not all, discrepancies between yeast growth and clinical data. We emphasize that the power of an allelic series lies in the diversity of phenotypes, which derive from distinct protein functions and reveal allele classes that may respond differently to treatment.

The full set of alleles demonstrated that mutation of any CBS domain could abrogate function, and remediation was not specific to cofactor-binding residues (Figure 7). B6 and heme sites are separated in the tertiary structure of CBS (Meier et al. 2001), yet some variants were remedial by either cofactor. Dual remedial alleles favored a global mechanism for cofactor rescue over the simpler model that increasing the cofactor concentration overcomes mutations that decrease the Km of cofactor binding (Ames et al. 2002; Wittung-Stafshede 2002). Since many characterized disease-causing mutations alter protein function via folding/stability (Yue et al. 2005; Kozich et al. 2010), alleles encoding unstable proteins may benefit from the binding energy provided by protein–cofactor interaction. Rescue of CBS function by biological or chemical chaperones is consistent with this hypothesis (Singh et al. 2007, 2010; Majtan et al. 2010).

Figure 7.

CBS phenotypes in relation to primary structure. Diagram of the domain structure of the CBS protein with the location of the 84 alleles used in this analysis represented by colored bars above the diagram. Each bar represents an allele; colors indicate the affect of the allele on growth. The Robust row reports the position of alleles indistinguishable from the predominant allele.

Regardless of the biochemical mechanism, cofactor availability regulated enzyme function for all CBS alleles within a narrow and physiologically relevant range of cofactor concentrations. While fully functional alleles supported growth at lower cofactor concentrations, metabolite levels of cells with a functional allele grown with a low B6 level and of cells with a nonfunctional allele were similar. This similarity of profiles may reflect a bone fide regulatory mechanism coupling pathway flux to nutrient availability. Similarly, substrate limitation affected trans-sulfuration flux as strongly as cofactor limitation. Methionine limitation reduced the level of homocystine in yeast cells regardless of whether CBS was functional or attenuated by limiting B6 (Figure 6). Indeed, since methionine catabolism leads to homocysteine formation, a low methionine diet is part of the treatment strategy for homocystinuria. Critically, glutathione production was also compromised by loss of CBS function or methionine limitation, with similar consequences to growth but different effects on homocystine production. Although patients with homocystinuria have relatively normal serum glutathione levels (Hargreaves et al. 2002; Orendac et al. 2003), tissue concentrations may be significantly lower (Maclean et al. 2010). Our data suggest that glutathione deficiency and homocysteine toxicity should be considered in evaluating the pathology of CBS deficiency.

Overall, inability to drive sufficient flux through the trans-sulfuration pathway, regardless of cause, led to growth defects (Figure 5 and Figure 6). Conventional thought about inborn errors is that metabolites accumulate at the point of the block. However, reversible reactions, circular connections, shunts in or out of a pathway, and feedback regulation, can establish new ratios among even distant metabolites. Thus, a more thorough understanding comes from parsing the symptoms as a function of alleles and related metabolites. For example, our quantitative assays revealed the subset of alleles that were more sensitive to B6 level and also provided evidence that the proteins encoded by six alleles benefited from increased heme level more than the predominant CBS allele. The behavior of these alleles suggested that heme deficiencies could complicate the diagnosis and treatment of homocystinuria. Conversely, the successful demonstration of heme supplementation could have utility in the clinic, either in addition to current treatments or as a second treatment formulation for certain alleles.