Megaloblastic anemia (MA) describes a heterogeneous set of anemias characterized by the presence of megaloblasts (immature red blood cells) in the bone marrow. MA develops as a result of ineffective DNA synthesis in rapidly dividing hematopoietic cells (1). Major causes of MA include vitamin deficiencies and exposure to certain drugs (2), termed drug-induced megaloblastic anemia (DIMA). As nutritional deficiencies have decreased in Western populations, the percentage of MA caused by drugs (i.e., DIMA) has increased (2, 3). DIMA occurs through various mechanisms, some of which remain elusive (Fig. 1).

Fig. 1
figure 1

Common mechanisms underlying megaloblastic anemia and drug-induced megaloblastic anemia. Major causes of megaloblastic anemia (MA) include deficiencies of vitamin B12 and folic acid due to inadequate dietary intake or malabsorption. MA due to drug exposure is also known as drug-induced megaloblastic anemia (DIMA). DIMA can be caused by a variety of drugs that inhibit DNA synthesis at different points of the DNA synthesis pathway. a Drugs can cause MA by reducing cellular availability of vitamin B12 or folic acid through reduced absorption, plasma transport, or delivery of folate or vitamin B12 or physical destruction of the vitamins (mechanisms #1 and #2). b Additional mechanisms of DIMA include inhibition of key enzymes such as dihydrofolate reductase resulting in tetrahydrofolate deficiency (mechanism #3) and drugs acting as purine and pyrimidine antagonists or analogs, including chemotherapies (5-fluorouracil), immune antagonists (e.g., leflunomide), and antiviral agents (mechanism #4). Interestingly, some drugs cause MA through an unknown mechanism. We propose an unknown mechanism of DIMA may involve inhibition of SLC19A2 in hematopoietic cells (mechanism #5)

Thiamine deficiency can be caused by poor dietary intake, chronic alcoholism, poor absorption (e.g., gastric bypass surgery or diarrhea), increased utilization (e.g., pregnancy), genetic mutations, and use of particular medications (e.g., diuretics) (4). Thiamine deficiency usually occurs systemically but it can occur locally in specific tissue/organs. Systemic thiamine deficiency can cause disorders such as beriberi and Wernicke-Korsakoff syndrome, whereas tissue/organ-specific thiamine deficiency due to genetic inactivation of thiamine transporters can lead to thiamine metabolism dysfunction syndrome-2 (THMD2, OMIM# 607483) or thiamine-responsive megaloblastic anemia (TRMA, OMIM# 249270) (4).

TRMA is characterized by three hallmark signs: MA, non-autoimmune type 1 diabetes, and sensorineural hearing loss (5). TRMA is caused by loss-of-function genetic mutations in SLC19A2, encoding the thiamine human transporter 1 (THTR1, SLC19A2) (6). SLC19A2 is the main thiamine transporter in pancreatic islet tissue and hematopoietic cells (7, 8). Approximately 60 SLC19A2 variants have been associated with TRMA (Supplemental Table 1) (5, 9), yet few have been functionally characterized to date. A mouse model of dietary thiamine deficiency (TD) has been used extensively to evaluate the effects of systemic thiamine depletion on neurological functions (10). However, there are no studies employing the TD model (or any in vivo system) to evaluate the impact of systemic thiamine depletion on erythropoiesis and megaloblastic anemia. Further, no studies have explored the idea that pharmacologic inhibitors of thiamine transporters may be causal for megaloblastic anemia (4). To our knowledge, SLC19A2 inhibition, which may phenocopy loss-of-function SLC19A2 variants, is not a known mechanism for DIMA.

In this study, we tested the hypothesis that drugs associated with DIMA may act (at least in part) as inhibitors of SLC19A2, and phenocopy genetic variants in the transporter. The objectives of our study were to (1) functionally characterize selected SLC19A2 variants associated with TRMA and (2) identify inhibitors of SLC19A2 among drugs associated with DIMA. The current study extends our work on thiamine transport mechanisms with a focus on THTR1 (SLC19A2). In particular, we functionally characterize several genetic mutations in SLC19A2 and identify clinically relevant SLC19A2 inhibitors. Collectively, our studies expand our understanding of drug-induced thiamine disorders.


Chemicals and Reagents

[3H]-Thiamine hydrochloride (catalog #ART 0710) was purchased from American Radiolabeled Chemicals Incorporation (St. Louis, MO, USA). Non-radiolabeled compounds were purchased from Sigma-Aldrich, Inc. (USA), Selleck Chemicals (Houston, TX), VWR International, Inc., and Thermo Fisher Scientific (USA). Cell culture supplies were purchased from Thermo Fisher Scientific (USA).

Functional Studies

Plasmids encoding wild-type SLC19A2 and three TRMA-associated variants were generated by Genscript Inc. (Piscataway, USA). Methods for generation of stably overexpressing SLC19A2 cell lines and uptake studies have been previously described (11, 12). Inhibition and kinetic studies were performed in SLC19A2-overexpressing stable HEK293 cells transiently transfected with an additional 500 ng of SLC19A2 wild-type plasmid.

Confocal Microscopy of SLC19A2 GFP-Tagged Variants

The subcellular localization of wild-type and variant SLC19A2 was determined with confocal microscopy on a Nikon CSU-22 Spinning Disk. Samples were prepared as previously described (12).

Inhibitor Screening Studies

Prescription drugs associated with MA (2) in addition to others belonging to listed drug classes (e.g., proton-pump inhibitors; 63 compounds total) were screened in triplicate at 1 mM to evaluate inhibitory activity against SLC19A2. The screen was performed using [3H]-thiamine (25 nM). Each plate contained control wells exposed for 10 min (within linear range, Supplemental Figure 1) to [3H]-thiamine (positive control), [3H]-thiamine plus 1 mM of amprolium (negative control), and [3H]-thiamine plus 1 mM of screened drug (3 replicates per drug). Methods have been previously described (11). SLC19A2 inhibitors were defined as compounds that decreased SLC19A2-mediated thiamine uptake by 50% or more at 1 mM. For compounds that met this definition, inhibition assays were conducted and IC50 values were calculated by nonlinear fitting using GraphPad Prism 7 (La Jolla, CA) (11).

Calculation of SLC19A2 Inhibitor Systemic Blood Concentrations

Clinical pharmacokinetic data from human studies (13,14,15,16,17,18) and drug monograph databases (e.g., Micromedex, FDA) were used to obtain Cmax (maximum plasma concentration) and fu (fraction of drug unbound in the plasma). Comparisons between unbound Cmax (fu*Cmax) and IC50 obtained in vitro ((fu*Cmax)/IC50 > 0.1) were used to estimate the likelihood that a drug may inhibit SLC19A2 clinically (Table I) (19, 20).

Table I Predicted Risk of Causing an SLC19A2-Mediated Clinical Drug-Drug Interaction Using the FDA Criteria (19, 20)

EHR Data Analyses

Methods detailing electronic health record (EHR) data extraction, filtering, and analysis are listed in Supplemental Table 2 and Supplemental Table 3.

TPK1 Enzyme Assay

Methods detailing thiamine pyrophosphokinase 1 (TPK1) enzyme assay development and execution are listed in Supplemental Figure 2 legend.


SLC19A2 Variants Causal for TRMA Exhibit Complete Loss-of-Function

HEK293 cells transiently transfected with three previously uncharacterized TRMA-associated SLC19A2 variants (p.T170P, p.G172R, and p.G334D) had significantly reduced thiamine transport activity compared to cells transfected with wild-type SLC19A2 (one-way ANOVA p-value < 0.0001, Fig. 2). The activities of variants p.T170P and p.G172R were not significantly different from cells transfected with empty vector (Fig. 2a; Student’s t-test p = 0.052 and 0.49, respectively), while variant p.G334D retained minimal but significant thiamine transport (p =0.0009). Similarly, inhibition of SLC19A2 with 200 μM amprolium or trimethoprim (a commonly used antibiotic and SLC19A2 inhibitor) abolished wild-type SLC19A2 activity to levels comparable to the TRMA-associated variants (Fig. 2a).

Fig. 2
figure 2

Function and expression of TRMA-associated SLC19A2 variants, and the effects of two prescription drugs on SLC19A2-mediated thiamine uptake. HEK293 cells transiently transfected with TRMA-associated SLC19A2 variants (p.T170P, p.G172R, and p.G334D) had significantly reduced thiamine transport compared to cells transfected with wild-type SLC19A2 transporter (one-way ANOVA p-value < 0.0001); functions of SLC19A2 variants were similar to that of empty vector cells (Student’s t-test EV versus p.G172R, p-value = 0.49). At 200 μM, the known SLC19A2 inhibitors trimethoprim and amprolium abolished thiamine transport by wild-type SLC19A2 resulting in transporter function comparable to the TRMA variants. Figures representative of 3 separate experiments (a). Membrane localization of GFP-tagged SLC19A2 variants. To determine TRMA variant membrane expression, constructs encoding SLC19A2 wild-type and TRMA variants were GFP-tagged and transfected in HEK293 cells. Nuclei were stained with Hoechst 33342 dye (blue) and cells were visualized with confocal microscopy (b)

SLC19A2 Variants Exhibit Disrupted Plasma Membrane Localization

Confocal microscopy of HEK293 cells expressing GFP-tagged wild-type and variant SLC19A2 revealed that the wild-type SLC19A2 localized primarily to the plasma membrane of the cell (Fig. 2b). In contrast, all three SLC19A2 variants exhibited partially disrupted membrane localization, and displayed concentrated puncta consistent with retention of the transporter variants in membrane-bound organelles.

Inhibitor Screen Identifies Ten Drugs That Substantially Inhibit SLC19A2

Among the 63 drugs screened for SLC19A2 inhibition (Fig. 3, Fig. 4, Supplemental Table 4), ten were designated as inhibitors, that is, these 10 substantially inhibited SLC19A2 (≥ 50% reduction in thiamine transport) at 1 mM. Of the ten inhibitors, four (fedratinib, amiloride, trimethoprim, amprolium) were previously identified as SLC19A2 inhibitors (11, 20), whereas six were novel SLC19A2 inhibitors, including erythromycin, mycophenolate mofetil, omeprazole, pantoprazole, lansoprazole, and chloroquine.

Fig. 3
figure 3

Common mechanisms of drug-induced megaloblastic anemia and SLC19A2 inhibitor drug screen workflow. A list of drugs associated with inducing megaloblastic anemia through various mechanisms was compiled (a). Each drug was dissolved to a final concentration of 1 mM in buffer containing radiolabeled [3H] thiamine and exposed to HEK293 cells expressing SLC19A2 wild-type transporter for 10 min. Drugs that inhibited 50% or more of SLC19A2-mediated thiamine uptake were considered significant inhibitors (10 total). Next, IC50 experiments were conducted for the top six inhibitors in the screen. Last, we selected SLC19A2 inhibitors for EHR analysis that reached the following threshold: (fu*Cmax)/(IC50) > 0.1 (b)

Fig. 4
figure 4

Drug screen and potency studies against SLC19A2-mediated thiamine uptake. In the screen, SLC19A2 inhibitors were defined as compounds that inhibit at least 50% of thiamine uptake compared to the positive control. The positive control was thiamine uptake in the absence of inhibitor. Out of the 63 drugs screened, ten reached that threshold. Of the ten drugs, three (fedratinib, amprolium, and trimethoprim) have IC50 values published in literature (a). IC50 experiments were conducted using inhibitor concentrations from 0 to 1 mM. IC50 curves for erythromycin, omeprazole, amiloride, mycophenolate mofetil, chloroquine, and pantoprazole were generated by fitting data to a nonlinear model (b). Figures representative of 3 independent experiments except for pantoprazole

IC50 Assays Reveal Erythromycin Potently Inhibits SLC19A2

Out of all the newly identified SLC19A2 inhibitors, erythromycin was the most potent with an estimated IC50 of 20 μM. Inhibition potency was followed by amiloride (69 μM), mycophenolate mofetil (145 μM), omeprazole (207 μM), chloroquine (301 μM), and pantoprazole (429 μM) (Table I).

Erythromycin Reaches Systemic Concentrations Relevant for SLC19A2 Inhibition

Calculation of unbound Cmax/IC50 values for the SLC19A2 inhibitors listed in Table I revealed that erythromycin was the only drug that met the FDA’s criteria for clinically relevant transporter inhibition.

Erythromycin Is a Substrate of TPK1

Luminescent signals from each enzyme reaction were measured and compared for omeprazole, erythromycin, thiamine (positive control or PC), and no substrate (negative control or NC). Luminescent signal in the erythromycin reaction was significantly higher than the negative control signal (Supplemental Figure 2, Student’s t-test, p-value < 0.0001), but still lower than the signal from the positive control (thiamine 1 mM). For omeprazole, the signal was statistically similar to the negative control (Supplemental Figure 2, Student’s t-test, p-value = 0.72).


In our previous research, we discovered that thiamine is the major endogenous substrate of OCT1 transporters (21), identified OCT1 as a key determinant of certain cardiometabolic (22), and hepatic (23) traits and identified several prescription drugs that are clinically relevant drug inhibitors of the intestinal thiamine transporter, SLC19A3 (24). In this study, we expand on our previous research by evaluating SLC19A2 expression and function and the physiological effects of SLC19A2 inhibition.

Approximately half of the 60 SLC19A2 mutations associated with TRMA are missense variants (Supplemental Table 1) (5, 25), yet few have been functionally characterized. In this study, we determined that three previously uncharacterized TRMA-associated SLC19A2 missense variants exhibited near-complete loss-of-function (Fig. 2). Consistent with loss-of-function, confocal microscopy revealed that the SLC19A2 variants largely failed to traffic to the plasma membrane. Instead, intracellular puncta were observed for all three variants, suggestive of impaired transporter trafficking or recycling, a pattern that has been observed previously for mutations in SLC19A2 (26, 27).

In the inhibitor screen, three prescription drugs associated with DIMA had IC50 values of < 200 μM. However, only the macrolide antibiotic, erythromycin, was predicted to inhibit SLC19A2 at clinically relevant concentrations when applying the FDA criteria for transporter-mediated drug-drug interactions. Erythromycin-induced MA is thought to occur through inhibition of folic acid absorption (2). However, our data suggest an additional mechanism via SLC19A2 inhibition (Fig. 4b), and potentially from inhibition of TPK1 (the enzyme that converts thiamine to TPP, the bioactive form of thiamine) (Supplemental Figure 2). Our finding that erythromycin is a substrate of TPK1 (Supplemental Figure 2) together with the fact that the drug is known to accumulate intracellularly to high levels (28) suggests that inhibition of TPK1 as well as of SLC19A2 may contribute to low TPP levels and ultimately to erythromycin-induced MA. EHR data were consistent with reduced levels of TPP in patients treated with erythromycin (Fig. 5, Supplemental Table 2, Supplemental Table 3). This mechanism may be particularly important in patients who may have higher systemic levels of erythromycin (e.g., patients with liver disease, the elderly, CYP3A4 poor metabolizers). Today, erythromycin use is rare compared to other macrolides (e.g., azithromycin).

Fig. 5
figure 5

Methods used to extract information from electronic health records (EHR) and results of thiamine pyrophosphate levels in patients prescribed erythromycin. To explore whether drugs predicted to cause a clinical drug-drug interaction in in vitro studies inhibit SLC19A2 in vivo, we extracted and analyzed data from the UCSF EHR. Data was filtered by presence (“on”) or absence (“off”) of erythromycin prescription, medication order start date, and thiamine pyrophosphate laboratory data. Before analysis, individuals in the “on” and “off” group were age and sex-matched (a). The thiamine pyrophosphate levels of patients prescribed erythromycin (+, N = 4) versus not prescribed erythromycin (−, N = 200) were compared. The average TPP levels for patients prescribed erythromycin were lower (137 nM) than that of patients not prescribed erythromycin (186 nM) (Welch’s two sample t-test, p-value = 0.0016)


Three TRMA-associated SLC19A2 variants were characterized, revealing complete loss of thiamine transport at least partially attributed to disrupted membrane localization. Our data suggest that SLC19A2 inhibition is not the mechanism behind DIMA for most associated prescription drugs, although it may contribute to erythromycin-induced MA.