Impact of Hydroxycarbamide Treatment on the Whole Blood Transcriptome in Sickle Cell Disease

Abstract

Hydroxycarbamide (HC) is the most widely used therapeutic for individuals with sickle cell disease (SCD). HC’s clinical benefits are primarily associated with its ability to induce fetal hemoglobin (HbF); this limited view of HC’s therapeutic potential may lead to its discontinuation when a modest amount of HbF is induced. A better understanding of the HbF-independent effects of HC on genes and pathways relevant to SCD pathophysiology is therefore needed. In this study, we performed bulk RNA-Seq on whole blood samples collected from a cohort of 25 pediatric patients with SCD to identify genes and pathways that are affected by treatment with HC. At the maximum tolerated dose (MTD) of HC, patients showed altered expression levels of several genes and biological pathways. Pathways related to heme metabolism, interferon alpha response, and interferon gamma response were significantly downregulated at HC MTD relative to the matched pre-HC samples. Pathways linked with IL2-STAT5 signaling and TNFα signaling via NF-Kβ were observed to be upregulated at HC MTD. These results illustrate the range of effects exerted by HC during therapy for SCD and pave the way for an improved understanding of the HbF induction-independent benefits of HC.

Graphical Abstract

Hydroxycarbamide (HC) is the most widely used therapeutic for individuals with sickle cell disease (SCD). HC’s benefits are primarily associated with its ability to induce fetal hemoglobin (HbF); this limited view of HC’s therapeutic potential may lead to its discontinuation when a modest amount of HbF is induced. Here, we performed bulk RNA-Seq on whole blood samples collected from 25 pediatric patients with SCD to identify genes and pathways affected by treatment with HC. Pathways related to heme metabolism, interferon gamma response, and interferon alpha response were significantly downregulated at HC MTD relative to the matched pre-HC samples. Pathways linked with IL2-STAT5 signaling and TNFα signaling via NF-Kβ were observed to be upregulated at HC MTD. These results pave the way for an improved understanding of the HbF induction-independent benefits of HC.

INTRODUCTION

Sickle cell disease (SCD) is an inherited blood disorder affecting approximately 100,000 Americans(1) and millions globally(2), with the highest incidence in sub-Saharan Africa(3). The causative beta globin mutation produces abnormal hemoglobin that polymerizes under hypoxia, resulting in sickle-shaped erythrocytes that can disrupt blood flow and cause pain events, organ damage, and early mortality(4). Clinical complications associated with SCD include bacterial infections, stroke, chronic kidney disease, and pulmonary hypertension.

Hydroxycarbamide (HC, also known as hydroxyurea) was the first drug approved for SCD treatment in adults and received FDA approval for use in children in 2017(5); The National Heart Lung and Blood Institute (NHLBI) guidelines recommend initiating HC therapy for SCD at nine months of age. HC induces fetal hemoglobin (HbF), which, at sufficient levels, can ameliorate the hypoxia-induced polymerization of sickle hemoglobin(6). Several signaling pathways have been proposed in HbF induction by HC; these include the nitric oxide-cyclic guanosine monophosphate pathway, epigenetic modification of the gamma-globin genes that encode HbF subunits, and the mitogen-activated protein kinase pathway(7). Similarly, microarray(8) and genome-wide association studies (GWAS) have uncovered variants associated with HbF induction after treatment with HC, leading to the identification of genes that encode HbF repressors such as BCL11A and HBS1L-MYB⁹. HbF (α₂ϒ₂) is the dominant form of hemoglobin produced in newborns and children up to two years of age, after which there is a shift to adult hemoglobin (α₂β₂); the decline of HbF levels marks the onset of clinical symptoms of SCD.

The focus on HbF induction has led some providers to view HC therapy as futile if significant induction is not experienced by an individual (9). However, there are numerous significant clinical benefits of HC therapy apart from HbF induction, such as the elevation of the mean corpuscular volume (MCV), which improves erythrocyte hydration and decreases erythrocyte density, and reduction in both white cell counts and endothelial adhesion(5,10). Although its efficacy as a therapeutic for SCD has been established, the molecular mechanisms by which HC exerts its multiple effects on SCD remain unclear. Despite recent efforts, we do not have a comprehensive picture of all the genes and biological pathways impacted by treatment with HC. Characterizing the gene expression changes will enable us to shed light on the non-HbF benefits of HC, particularly those not captured by conventional laboratory testing such as complete blood counts. With this study, we aimed to investigate the effect of HC on pediatric patients with SCD using a whole blood bulk transcriptomics approach.

METHODS

Study Population

The study population included 25 pediatric patients with sickle cell anemia (SCA) aged between 1 and 19 years (mean age: 9 years). The demographics and lab measurements of the 25 patients are presented in Table 1. Two individuals in the cohort were compound heterozygotes, with HbSβ⁰ thalassemia; the rest were HbSS. Hematologic lab measurements (including %HbF) were obtained by chart review pre-HC and at the maximum tolerated dose of hydroxycarbamide (HC MTD). All patients reached MTD on HC therapy within 10 months and for 23 of them, matched pre-HC and HC MTD samples were collected. For the remaining two patients, pre-HC samples were not obtained, but samples were collected at HC MTD and after at least three months of hydroxycarbamide discontinuation (henceforth known as “washout samples”). The mean duration of hydroxycarbamide treatment was 9.5 months. All samples were obtained under informed consent. The study protocol was approved by the Baylor College of Medicine Internal Review Board.

Table 1:

Patient cohort summary

	Baseline (n=23)	HC at MTD (n=25)	HC washout (n=2)*	Significance test based on paired baseline vs. HC at MTD (n=23)
Sex: Male, n	13	13	0
Sex: Female, n	10	12	2
Age, mean (SD), years	7.66 (6.34)	8.94 (6.44)	13.32, 18.48b	-
Genotype: HbSS, n	21	23	2
Genotype: HbSβ0, n	2	2	0
HC treatment duration, mean (SD), years		0.79 (0.39)	0.75, 1.36	-
Mean %HbF†	16.05 (11.56)	21.31 (11.63)	10.39, 9.86	0.0001
Mean Delta %HbF‡		5.46 (5.91)d	12.88, 4.74	-
Mean Hb, g/dL	8.49 (1.18)	9.48 (1.44)	8.80, 8.20	0.0001
Mean MCV, fL	83.80 (9.84)	98.57 (11.61)	88.70, 95.90	<0.0001
Mean WBC, ×1000/μL	13.00 (4.04)	7.54 (1.82)	10.43, 8.96	<0.0001
Mean ANC, /μL	5525.21 (2708.27)	2827.20 (1160.91)	4480.00, 4420.00	0.0001
Mean ARC, ×1000/μL	505.99 (216.24)	323.76 (104.15)	446.30, 450.30	0.0004

^*Entries for age and the hematological parameters in this column represent individual values.

^†Corrected %HbF=%HbF/(%HbF+%HbS) as determined by HPLC.

^dDelta %HbF = Corrected %HbF at MTD – corrected %HbF at baseline/washout.

ANC, absolute neutrophil count; ARC, absolute reticulocyte count; Hb, hemoglobin; HbF, fetal hemoglobin; HbSβ⁰, sickle beta zero thalassemia; HbSS, homozygous for SCD; HPLC, high performance liquid chromatography; HC, hydroxycarbamide; MCV, mean corpuscular volume; MTD, maximum tolerated dose; SD, standard deviation; WBC, white blood cell count.

Gene Expression Profiling

A total of 3 mL of whole blood was collected in EDTA tubes for each patient during either baseline or HC washout, and at HC MTD and processed within four hours of collection. The total RNA was isolated from the samples without a globin-depletion step and cDNA was sequenced on an Illumina HiSeq 4000 instrument (Illumina, San Diego, CA) with 150 bp paired-end, un-stranded library preparation protocol according to the manufacturer’s instructions.

Quality control of the paired-end reads was performed using FastQC (version 0.11.9)(11). Since adapter contamination in the reads was less than 5%, the input reads were not further trimmed. Read alignment was performed using STAR (version 2.7.10b)(12) with the GRCh38 primary assembly and GENCODE annotation (release 42)(13). The number of reads per gene was quantified using HTSeq (version 2.0.2) with the same GENCODE annotation as reference.

Genes with zero expression in at least 10% of the samples were removed, and the raw read counts were normalized for library size with the trimmed mean of M values (TMM) method implemented in the edgeR package(14). The R package PVCA was used to examine sources of variability in gene expression(15). Patient sex was verified using the expression of the sex-specific genes RPS4Y1 and XIST. The normalized counts were subsequently log-transformed and provided as input to Supervised Normalization of Microarrays (SNM)(16) for the adjustment of age, sex, absolute neutrophil counts, absolute reticulocyte counts, and white blood cell counts as covariates. Age was treated as a categorical variable, with samples being categorized as either above or below 5 years of age. The output gene expression matrix from SNM was utilized for a paired differential gene expression analysis (DGE) with the limma-trend pipeline(17) using the duplicateCorrelation function to model subject as a random effect. Genes with Benjamini-Hochberg adjusted p-values < 0.05 were considered to be differentially expressed between baseline and HC MTD.

The log-fold changes and adjusted p-values for each gene were provided as input to the fgsea R package(18) with the MSigDB(19,20) database as the reference for identifying enriched gene sets/biological pathways; the ranking metric used was the product of the negative logarithm of the p-value multiplied by the log fold-change. Pathways with adjusted p-value < 0.05 were considered significantly enriched. Complete code for the DGE analysis is available on our GitHub repository: https://github.com/vtbhat/SCD_Omics_Scripts.

Statistical Analysis

For the hematologic lab data shown in Table 1, we used the R package nlme(21) to implement linear mixed models with HC treatment status as the predictor, adjusting for age, sex, and genotype as fixed effect variables, and subject ID and treatment duration as random effect variables. The principal component analysis was performed using PCAtools(22).

The R package glmmSeq was used to perform ANCOVA and determine changes in gene expression changes that are associated with HbF induction by HC. The age-adjusted gene expression values for 22 subjects (that is, excluding the washout samples and one subject that showed a decrease in absolute %HbF at HC MTD) were provided as input, with treatment status as fixed effect, log %HbF levels as a continuous covariate, and subject ID as random effect. Since HbF levels generally decline after two years of age in individuals with SCD(23) (a trend also noted in this dataset), it is difficult to quantify the rise in HbF due to HC in individuals who started on the drug before the age of 2, as the drug effect is masked by the physiologic decline. Hence, we categorized the samples as either below two or above two years for the adjustment of age in SNM.

The genes that passed the Benjamini-Hochberg threshold of 0.05 for both treatment status and HbF levels as predictors were retained for over-representation analysis. Over-representation analysis was performed using clusterProfiler(24,25) and the Reactome(26) database as the reference to decipher the gene signatures and biological pathways that they may potentially overlap with. Gene signatures with a Benjamini-Hochberg FDR < 0.05 were noted.

RESULTS

Comparison of HC MTD to HC washout or baseline HC

Compared with baseline, there was a robust increase in %HbF levels at MTD which was accompanied by a significant increase in hemoglobin (Hb) and MCV and a decrease in white blood cell count (WBC), absolute neutrophil count (ANC), and absolute reticulocyte count (ARC)(Table 1). Following HC washout, the %HbF, MCV, and ARC returned to levels comparable to those at baseline. However, ANC (per μL) remained comparable to HC MTD (HC MTD: patient 1=3700, patient 2=4290; HC washout: patient 1=4480, patient 2=4420). The effect of age on %HbF and MCV was significant at baseline (P=0.00019 for %HbF, P= 0.02 for MCV) and at HC MTD (P=0.00041 for %HbF, P=0.00077 for MCV) in a linear model with age as predictor and sex as covariate. The effect of age on ARC and ANC was not significant at either baseline or HC MTD.

In order to determine how the washout samples differed from the remaining samples collected at baseline or HC MTD, we performed a principal component analysis (PCA) with the log-transformed gene expression values of all samples. The PCA biplot (Figure 1) indicated that the two washout samples clustered with the HC MTD samples and were relatively close to their matched sample, implying that the two individuals may have not reverted to a pre-HC state after HC discontinuation despite return to baseline hematologic indices. The first two principal components captured approximately 35% of the total variation in the dataset. Figure 2 shows the proportion of variation in gene expression explained by the first five PCs (50%) that can be attributed to age (older or younger than 2 years), sex, condition (baseline and HC MTD), and their interactions. All three of these main effects explained between 5 and 10% of the variance, without meaningful interactions.

Figure 1.

Principal component analysis biplot: PCA biplot depicting the clustering of the pre-HC, HC MTD, and HC washout samples. The gene expression has been adjusted for sex and age. The washout samples (in green) clustered with the HC MTD samples (in red).

Figure 2:

Principal variance component analysis: PVCA barplot shows the proportion of variance of gene expression explained by each effect (age category, sex, and condition) and their interactions.

Effect of HC Treatment on Gene Expression

We performed whole blood gene expression profiling to analyze genes and pathways that are differentially expressed between baseline and HC MTD within a cohort of 23 pediatric SCD patients. The mean library size for RNA-Seq was approximately 20M reads per sample. After filtering out genes with low expression and TMM-based normalization for library sizes, 23,164 genes were retained. To control for differential blood cell abundance, analyses were performed after fitting neutrophil, white blood cell, and reticulocyte counts as covariates during normalization.

A total of 2124 genes with Benjamini-Hochberg p-values less than 0.05 and absolute log2-fold change greater than 0.5 were found to be differentially expressed between pre-HC and HC MTD, with 1563 genes showing elevated expression levels (log2-fold change > 0.5) and 561 showing decreased expression (log2-fold change < 0.5) at HC MTD (Supplementary Table 1). Pathway enrichment analysis with the DGE result matrix as input revealed several pathways being significantly enriched in response to HC (Figure 3). Pathways related to heme metabolism, interferon alpha response, and interferon gamma response were downregulated at HC MTD, while pathways linked to IL2-STAT5 signaling and TNFα signaling via NF-Kβ were observed to be significantly upregulated at HC MTD. The complete set of enriched pathways and their associated statistics is described in Supplementary Table 2. Similar results were observed without adjustment for blood cell counts.

Figure 3:

Pathway enrichment analysis results: Barplot depicting the pathways upregulated or downregulated at HC MTD when compared to the baseline. Pathways with positive enrichment scores are up-regulated in HC MTD samples, with darker shading indicating significance after adjustment for the number of comparisons.

Impact of HC on HbF Induction

Although our main goal was to elucidate the HbF independent effects of HC, given the intense interest in HbF regulation, we have also characterized expression changes that accompany HbF induction by HC using ANCOVA. ANCOVA analysis with the %HbF values and the gene expression values showed that after correction for multiple testing, 461 genes had coefficients > 0 (adjusted p-value < 0.05, positive correlation), and 642 genes had coefficients < 0 (adjusted p-value < 0.05, negative correlation). A total of 252 genes exhibited different levels of expression between pre-HC and HC MTD and were correlated with %HbF levels (Benjamini-Hochberg p-value < 0.05); these genes may be involved in HC-induced HbF induction. The five genes with the strongest positive correlation and strongest negative correlation with log %HbF are listed in Table 2 and are discussed further below. Multiple gene signatures were determined to be statistically significantly associated with the degree of HC induction of HbF (Table 3).

Table 2:

Top genes most strongly correlated with the log %HbF levels and associated with treatment status in the ANCOVA

Ensembl Gene ID	HGNC Symbol	Direction of correlation	P-value	Adjusted P-value
ENSG00000213934	HBG1	Positive	7.717E-04	2.233E-02
ENSG00000211679	IGLC3	Positive	2.518E-05	4.413E-03
ENSG00000196092	PAX1	Positive	8.168E-06	2.359E-03
ENSG00000105369	CD79A	Positive	5.175E-04	1.841E-02
ENSG00000132465	JCHAIN	Positive	1.134E-04	8.954E-03
ENSG00000137878	GCOM1	Negative	1.924E-06	9.957E-04
ENSG00000198892	SHISA4	Negative	2.199E-03	3.928E-02
ENSG00000054793	ATP9A	Negative	4.151E-08	9.517E-05
ENSG00000223688	RPS24P14	Negative	5.199E-08	1.022E-04
ENSG00000005882	PDK2	Negative	1.561E-03	3.283E-02

Table 3:

Overrepresentation analysis results: Pathways enriched in the genes associated with HbF induction by HC.

Signature	Correlation with HbF	P-Value	Adjusted P-value	Number of overlapping genes/Total genes in the pathway
Antigen activates B Cell Receptor (BCR) leading to generation of second messengers	Positive	4.833E-05	9.629E-03	6/89
FCERI mediated Ca+2 mobilization	Positive	1.064E-04	9.629E-03	5/64
FCERI mediated MAPK activation	Positive	9.830E-05	9.629E-03	6/101
CD22 mediated BCR regulation	Positive	1.038E-04	9.629E-03	6/102
Potential therapeutics for SARS	Positive	1.213E-04	9.629E-03	7/151
Scavenging of heme from plasma	Positive	3.244E-04	2.111E-02	5/81

Interestingly, we observed that two patients had extremely high HBG1 expression levels at pre-HC and HC MTD; to ensure that these four samples were not outliers due to variation in RNA-Seq library preparation and sample processing, we examined the clustering of these outliers relative to the other samples in a sample-sample correlation heatmap (Supplementary Figure 1). There were no discernible differences between these four samples and the remaining samples in the heatmap; moreover, the HBG2 levels for the four samples were within the expected range (Supplementary Figure 2). The gene expression profiles for baseline and post-HC in the sample-sample correlation heatmap were clustered together for one of these two individuals. This suggests that the extreme HBG1 levels were true biological responses in these two children.

DISCUSSION

HC has been the most widely used SCD therapy for more than two decades; however, much remains to be learned about its mechanistic impact. Here, we investigated the treatment effect of HC on the whole blood transcriptomes of 25 pediatric patients with SCD. While multiple GWAS, microarray, and molecular studies have been published in the past, this is to our knowledge the first study that examines the response to HC at the transcriptome level with bulk RNA-Seq. However, our study has two major limitations. The first is the absence of an additional transcriptomics dataset to replicate our results for the correlation analysis between gene expression and HbF levels; validating the results with a larger dataset could potentially form the basis of a regression model for predicting the extent of HbF elevation after HC therapy. Secondly, gene expression profiling was performed with whole blood samples, which comprise several cell types; hence, it is difficult to explore cell-type-specific sources of differential gene expression. Future studies with single-cell sequencing techniques will be necessary to understand the extent to which the gene expression changes reflect changes in cell-type abundance.

Our analysis confirmed some of the known genes and pathways regulated in response to HC. The gene PPP6C, which encodes the catalytic subunit of protein phosphatase 6 and is a known HbF repressor(27), showed reduced expression at HC MTD. Two candidate genes that were identified in previously published GWAS investigating HbF induction were also found to be differentially expressed in this cohort. The gene HBS1L, which encodes a protein with GTP-binding activity, had lower levels of expression at HC MTD, consistent with a report using erythroid cultures that HbF levels are negatively correlated with HBS1L expression(28). Conversely, BACH2 exhibited increased expression at HC MTD. In a recent multi-ancestry GWAS, HbF levels were found to be repressed with increased BACH2 expression, but the authors also observed a small delay in erythroid maturation(29). Although HC can delay erythroid maturation in a dose-dependent manner(30), Kato et al.(31) showed that the BACH transcription factors can promote erythroid differentiation.

The heme metabolism geneset, which comprises genes encoding proteins involved both in the synthesis and degradation of heme and in erythroid differentiation, was downregulated after treatment with HC. This may reflect the known erythroid maturation arrest caused by HC, and a reduction in heme synthesis as a result of reduced red cell turnover. We observed lower expression of genes linked to the interferon-alpha response pathway after HC therapy, implying that HC may alleviate the increased activation of interferon-alpha signaling observed in individuals with SCD not on disease modification therapy(32). Similarly, HC also repressed the interferon-gamma response pathway, suggesting a reduction in inflammation; previous studies have demonstrated the elevation of interferon gamma during steady state and vaso-occlusive crises(33). The upregulation of IL2-STAT5 signaling and NF-Kβ activation by HC has been observed in previously published studies not directly linked to SCD(34,35); the upregulation of IL2 signaling may modulate adaptive immunity and thus contribute to a reduction in acute pain episodes(34).

Expression levels of the genes HBG2 and HBG1, encoding the hemoglobin gamma subunits, exhibited the strongest positive correlation with HbF levels; HBG1 was also significantly associated with treatment status. Many of the gene expression changes that were positively correlated with HbF induction and associated with treatment status include those that encode immunoglobulins and transcription factors that are primarily expressed in B cells, such as PAX5, TCF4, and BACH2(36). The gene GATA1, encoding a transcription factor that can combine with BCL11A and SOX6 to silence γ-globin gene expression(37), was negatively correlated with HbF but was not associated with treatment status. As noted, single-cell RNAseq will be required to tease apart which cell types mediate the observed whole-blood correlations with HbF levels.

The results from the overrepresentation analysis comprised six signatures that were positively linked to HbF levels; most of the gene signatures were related to B-cell receptor signaling. This can be attributed to the fact that nearly all the genes that overlapped between the input and the Reactome signatures were immunoglobulins and comprised less than a quarter of the total number of genes in these signatures. However, one signature—scavenging of heme by plasma—appears to support HC’s role in reducing inflammation. The sickling of erythrocytes in SCD causes intravascular hemolysis and the release of free heme in plasma. Free heme can activate neutrophils and macrophages, which release pro-inflammatory mediators and subsequently drive inflammation and organ damage(38). Wiatr et al(39)previously demonstrated that immunoglobulins can exhibit anti-inflammatory activity via heme scavenging, and thus potentially lower vaso-occlusion.

In conclusion, HC therapy for SCD appears to offer numerous advantages through diverse biological mechanisms. Treatment with HC increases the production of HbF by regulating the expression of genes such as PPP6C and HBS1L that are associated with gamma globin chain synthesis. Furthermore, HC affects multiple pathways that are independent of HbF induction, including the reduction of heme metabolism and interferon response. HC can decrease SCD-related complications such as vaso-occlusion by increasing the rate of heme scavenging from plasma. Since individuals with low HbF induction from HC typically exhibit high rates of hemolysis releasing heme, they may benefit significantly from heme scavenging, for example. Taken together, these results strongly suggest that the continuation of HC therapy can be beneficial to patients with SCD irrespective of the extent of HbF induction.

Supplementary Material

Supinfo3

Supplementary Table S2: List of enriched pathways and their associated statistics

Supinfo2

Supplementary Table S1: List of differentially expressed genes between pre-HU and HU MTD

Data Availability Statement

The bulk RNA-Seq data generated for this study is available at GEO under accession number GSE254951.