Myeloablation followed by hematopoietic stem cell transplantation leads to long-term normalization of systemic sclerosis molecular signatures

Abstract

Objective:

In the randomized Scleroderma: Cyclophosphamide Or Transplantation (SCOT trial), myeloablation, followed by hematopoietic stem cell transplantation (HSCT), led to normalization of systemic sclerosis (SSc) peripheral blood gene cell (PBC) expression signature at the 26-month visit. Herein, we examined long-term molecular changes ensuing 54 months after randomization for individuals receiving an HSCT or 12 months of intravenous cyclophosphamide (CYC).

Method:

Global PBC transcript studies were performed at pretreatment baseline, 38 months, and 54 months post-randomization, as well as in healthy controls using Illumina HT-12 arrays.

Results:

Thirty (HSCT=19 and CYC=11) participants had 38-month and 26 (HSCT=16 and CYC=11) had 54-month samples available. In the paired comparison to baseline, a significant down-regulation of interferon modules and an up-regulation of cytotoxic/NK module were observed at 38-month and 54-month visits in the HSCT arm indicating a long-term normalization of baseline SSc gene expression signature. No differentially expressed modules were detected in the CYC arm.

In the comparison to healthy controls, 38-month visit samples in the HSCT arm showed an upregulation of B cell and plasmablast modules and a downregulation of myeloid and inflammation modules. Importantly, 54-month HSCT samples did not show any differentially expressed modules compared to healthy controls suggesting completion of immune reconstitution. Participants in the CYC arm continued to show an SSc transcript signature in comparison to controls at both time points.

Conclusion:

Paralleling the observed clinical benefit, HSCT leads to durable long-term normalization of the molecular signature in SSc with completion of immune resetting to 54 months post-HSCT.

Systemic sclerosis (SSc; scleroderma) is an autoimmune disorder characterized by immune activation, vascular injury, and fibrosis. Myeloablation/immune ablation followed by autologous hematopoietic stem-cell transplantation (HSCT) shows efficacy for treatment of severe SSc (1-3). In the Scleroderma: Cyclophosphamide Or Transplantation (SCOT) trial, participants with SSc underwent either myeloablation followed by CD34+ selected autologous HSCT or 12 months of intravenous monthly cyclophosphamide (CYC) (3). The primary trial period was 54 months. Long-term outcomes, including event-free survival, were superior in the HSCT compared to the CYC group at that time point.

We previously reported peripheral blood cell (PBC) global gene expression changes post-HSCT showing transplantation, contrary to CYC treatment, normalizes SSc signature at 26 months (4). Specifically, an upregulation of the interferon (IFN) and neutrophil modules and a decreased cytotoxic/NK module were observed in the comparison of baseline SSc to healthy control samples. In the paired longitudinal comparison, the upregulated IFN and neutrophil modules at the baseline visit decreased significantly after HSCT, while the down-regulated cytotoxic/NK module at the baseline visit significantly increased after HSCT at the 26-month visit. Given these significant changes at 26-months post-transplantation, we herein investigate changes in PBC gene expression profiles in samples collected at 38 months or 54 months. We hypothesized that the signature for CYC-treated participants would be largely unchanged relative to baseline suggesting failure of this treatment to produce long-term improvements in the SSc immune profile. On the other hand, we hypothesized that the signature in the HSCT participants would, over the course of 54 months, more closely resemble healthy controls suggesting durable long-term normalization of the SSc molecular signature.

METHODS:

All SCOT participants met the 1980 American College of Rheumatology (formerly the American Rheumatism Association) preliminary classification criteria for SSc (5). They also fulfilled the following inclusion criteria: Diffuse cutaneous involvement (6); age 18-69 years; disease onset within the previous 5 years (defined as the first non-Raynaud's phenomenon symptom); early internal organ involvement with either pulmonary disease (diffusing capacity of the lung for carbon monoxide [DLCO] or Forced Vital Capacity [FVC] <70%) or prior scleroderma renal crisis. A detailed description of inclusion and exclusion criteria has been previously published (3).

Whole blood samples were collected in Tempus tubes (Applied Biosystems, Foster City, California, USA) and stored at −80°C in a central biorepository. RNA was extracted according to the manufacturer’s protocol. RNA quality was assessed using Bioanalyzer (Agilent Genomics, Santa Clara, California, USA) and those with RNA integrity numbers (RIN) >7 were included in the study. Similar to the 26-month follow-up study, global gene expression profiling was performed using Illumina HT-12 arrays (Illumina, San Diego, California, USA) (4). The laboratory personnel who conducted the molecular assays were blinded to disease status and treatment group assignment. The baseline and follow-up visit, as well as healthy control PBC gene expression data were normalized and filtered in one dataset. The data underwent quantile normalization followed by batch correction before being incorporated into the model. Traditional modular analysis using 62 curated whole blood modules was conducted using the original repertoire analysis as previously described (4,7). In this analysis, sets of co-expressed genes (modules) that are observed in whole blood across a variety of inflammatory and infectious diseases were investigated. If possible, a biological function is assigned to a module based on the function of genes present in the gene set (e.g., IFN, B-cell, plasmablasts, etc.), and these modules are called annotated modules. Other modules remained uncategorized (unannotated). In addition to traditional modular analysis, a gene set analysis was conducted using the QuSAGE algorithm for the modular analysis of differentially expressed genes (8). QuSAGE tests whether the average log2 fold change of a gene set is different from zero. The method correctly adjusts for gene-to-gene correlations within a gene set and provides an easily interpretable metric for the magnitude of differential regulation.

These analyses were performed using subsets of individuals in the per-protocol population (i.e., who received either a transplant or completed ≥9 doses of CYC) who had available PBC RNA samples with RIN>7 at analysis time points at the examined time points (Supplementary Figure 1). Sixty-two unaffected controls of similar age-, gender-, and racial background who were included in the 26-month SCOT follow-up study were also analyzed (4). The unpaired analyses of SSc samples to unaffected controls included 30 SSc samples (HSCT=19 and CYC=11) at 38-month and 26 SSc samples (HSCT=16 and CYC=11) at 54-months.Among those with available PBC gene expression data at 38-month and 54-month visits, only a small number of participants was treated with disease-modifying antirheumatic drugs in both treatment arms (Supplementary Figure 1). The paired analyses were conducted to evaluate changes in modular gene expression from baseline to months 38 and 54 in the subset of SCOT. All analyses were adjusted for age and gender. The criterion to identify differentially expressed modules was log2 fold change >0.25 or < - 0.25 and p_FDR <0.1, the same criterion used in the 26-month follow-up study (4). The gene expression data will be deposited on Gene Expression Omnibus.

We have previously shown that SCOT participants at the pretreatment baseline visit display an up-regulation of two IFN modules (M1.2 and M3.4) and one neutrophil module (M5.15) and a down-regulation of the cytotoxic/NK (M3.6) module in comparison to healthy controls (4). We investigated the possibility of an association between baseline gene expression, measured as transcript module scores, and the expected longitudinal course of the modified Rodnan Skin Score (mRSS) and FVC % predicted (FVC%) up to 72 months post-randomization, as previously described (9). Separate models were fit for each gene module. For FVC%, mixed effects regression models were fit as splines with separate intercepts and slopes for each treatment along with interaction terms to allow for modification by baseline gene expression. For HSCT, the spline included pivot points at 3 and 14-months post-transplant to allow for the expected to fall and recovery after total body irradiation. For CYC, the spline included a single pivot at 14 months. Subject-level effects for intercept, initial slope and post month-14 slope were included as random effects assuming separate banded (2)-unstructured covariance structures for each treatment. For mRSS, because the rate of decline decreases over time, a negative exponential decay model was used to estimate longitudinal trends. The nonlinear mixed model included separate fixed intercepts and decay constants for each treatment arm and interaction terms to allow for modification by baseline gene expression. Random subject-level effects for intercept and decay constant were included assuming separate unstructured covariance structures for each treatment. Sex, prior cyclophosphamide use, and smoking status were included as fixed covariates in all models. Because longitudinal data are missing for individuals who died or experienced organ failure, models should generally account for informatively missing data. In previous analyses, we compared treatment effects using mixed effects regression models, under the missing-at-random assumption, with shared parameter models, assuming data were not-missing-at-random (9). Results were consistent with more conservative treatment differences for the mixed effect regression models. Hence, for analyses present here, we assumed data were missing-at-random and included all available data for the per protocol population in each model.

RESULTS:

A baseline transcript profile was available in 58 participants (HSCT=26 and CYC=32) in the SCOT per-protocol population. Demographic and baseline characteristics are described in Supplementary Table 1.

Differential transcript module expression 38-months after treatment:

As mentioned previously, prior to treatment, two IFN modules [M1.2 and M3.4] and one neutrophil module [M5.15] were upregulated compared to normal controls, and the cytotoxic/NK [M3.6] module was downregulated in SSc (4). At 38 months in the HSCT arm, B-cell modules (B-cells, plasmablasts) and cytotoxic/NK cell modules were up-regulated, and IFN and innate immunity modules were down-regulated compared to baseline levels (Table 1, Figure 1A). Focusing on the aforementioned SSc signature, the IFN module [M1.2], which was the most prominent upregulated module at baseline, was significantly decreased (Figures 2A), while the down-regulated cytotoxic/NK cell module, M3.6, was significantly increased 38 months after transplant (Figure 2D). Compared to healthy controls, B-cell modules and cell cycle modules were up-regulated 38 months post-transplant, and innate immunity and erythropoiesis modules were down-regulated. The pre-treatment SSc signature (high IFN and neutrophil and low NK/cytotoxic modules) was not evident 38 months after transplant (Table 1 and Figure 1B).

Table 1.

Differentially expressed modules in comparison of SCOT 38- and 54-month to baseline samples in the HSCT arm or healthy controls based on QuSAGE analysis

Module	Annotation	Log2 Fold Change	Confidence Interval (95%)	Pvalue	PFDR
Month 38 compared to baseline
M4.11	Plasmablasts	0.521	(0.236,0.806)	<0.001	0.005
M3.6	Cytotoxic / NK Cell	0.469	(0.274,0.665)	<0.001	<0.001
M4.10	B cells	0.278	(0.068,0.488)	0.009	0.039
M4.15	Cytotoxic / NK Cell	0.251	(0.057,0.446)	0.011	0.042
M5.14	Myeloid Lineage	−0.269	(−0.445, −0.094)	0.003	0.021
M6.13	Inflammation	−0.288	(−0.472, −0.105)	0.002	0.021
M4.13	Inflammation	−0.307	(−0.539, −0.075)	0.010	0.039
M1.1	Coagulation / Platelets	−0.338	(−0.518, −0.158)	<0.001	0.005
M1.2	IFN Response	−0.515	(−0.796, −0.235)	<0.001	0.005
Month 38 compared to healthy controls
M4.11	Plasmablasts	0.597	(0.311,0.883)	<0.001	<0.001
M3.3	Cell Cycle / Proliferation	0.361	(0.213,0.509)	<0.001	<0.001
M4.10	B cells	0.262	(0.035,0.488)	0.024	0.047
M5.14	Myeloid Lineage	−0.265	(−0.424, −0.106)	0.001	0.005
M3.2	Myeloid Lineage	−0.268	(−0.441, −0.096)	0.002	0.007
M4.4	Erythropoiesis	−0.305	(−0.475, −0.136)	<0.001	0.003
M1.1	Coagulation / Platelets	−0.344	(−0.523, −0.164)	<0.001	0.002
M4.13	Inflammation	−0.353	(−0.554, −0.152)	0.001	0.003
M2.3	Erythropoiesis	−0.415	(−0.701, −0.129)	0.005	0.013
M3.1	Erythropoiesis	−0.535	(−0.802, −0.268)	<0.001	0.001
Month 54 compared to baseline
M2.3	Erythropoiesis	0.453	(0.162,0.744)	0.002	0.047
M3.1	Erythropoiesis	0.335	(0.071,0.598)	0.013	0.064
M3.6	Cytotoxic / NK Cell	0.291	(0.095,0.486)	0.004	0.055
M1.2	IFN Response	−0.474	(−0.749, −0.199)	0.001	0.046
Month 54 compared to healthy controls*

^*No modules passed criteria ((log2 fold change >0.25 or < - 0.25 and pFDR <0.1). HSCT = hematopoietic stem cell transplant

Figure 1.

Modular analyses of SCOT participant samples in the HSCT arm at 38 months and 54 months compared to baseline samples or healthy controls

A) Pairwise comparisons of HSCT 38-month to baseline samples. B) Comparisons of 38-month HSCT samples to healthy controls. C) Legend of color-coding in panels A, B, D and E demonstrating proportion of module transcripts over/under-expressed. D) Pairwise comparison of 54-month HSCT to baseline samples. E) Comparisons of 54-month HSCT samples to healthy controls. F) Annotation of modules based on known biological function of genes included in a given module. The numbers on Y- and X-axes indicate the main module and sub module designation, respectively. Of note, the module map in this Figure and results in Table 1 are based on two different analytic algorithms (repertoire analysis vs. QuSAGE). HSCT = hematopoietic stem cell transplant; CYC = cyclophosphamide.

Figure 2.

Composite scores of IFN [M1.2, M3.4], neutrophil [M5.15], and cytotoxic/NK cell [M3.6] modules in control and SCOT participants at baseline, 38-, and 54-months post-treatment*

Longitudinal measurements of A) M1.2 (IFN module), B) M3.4 (IFN module), C) M5.15 (Neutrophil module), and D) M3.6 (Cytotoxic/NK cell module). *Log2 fold change >0.25 and FDR<0.1 in paired analysis of follow-up to baseline SSc samples in QuSAGE analysis. The displayed data at all time points are restricted to the subgroup of participants who had PBC gene expression data available at the baseline and 38-month-, as well as 54-month visits, in order to compare the gene expression modules from the same individuals at these three-time points (sample size: HSCT arm:14- CYC arm: 8).

In contrast, in the CYC arm, no modules were differentially expressed at 38 months compared to baseline levels (Supplementary Table 2, Supplementary Figure 2A). Moreover, in comparison to healthy controls, the pre-treatment SSc signature was still present, as evident by a significant upregulation of IFN [M1.2 and M3.4] and neutrophil modules [M5.15] and a significantly decreased cytotoxic/NK cell (M3.6) module (Supplementary Table 2/Supplementary Figure 2B).

These 38-month findings parallel our previously reported 26-month data, indicating a “new normal” status in participants receiving HSCT, characterized by normalization of SSc baseline signature and downregulation of innate immunity-related inflammatory pathway, while B-cell-related modules were activated (4). In contrast, participants in the CYC continued showing a typical SSc transcriptomic signature.

Differential transcript module expression 54 months after treatment:

At month 54 in the HSCT arm, cytotoxic/NK cell and erythropoiesis modules were upregulated and an IFN module was down-regulated compared to baseline (Table 1/Figures 1D, 2A, and 2D). Furthermore, no modules were differentially expressed compared to normal controls, indicating that the HSCT 54-month samples are more similar to healthy controls than HSCT 38-month samples.

In the CYC arm, findings at 54 months are consistent with 38 months. No modules were differentially expressed at 54 months compared to baseline (Supplementary Table 2, Supplementary Figure 2D). Furthermore, in comparison to healthy controls, the pre-treatment SSc signature was still present at 54 months, as evident by an upregulation of IFN [M1.2 and M3.4] and neutrophil modules [M5.15] and a down-regulation of cytotoxic/NK cell modules [M3.6, M4.15] (Supplementary Table 2/Supplementary Figure 2E).

Effect of baseline SSc signature modules on the long-term clinical manifestations:

We investigated the possibility of an association between baseline transcript scores for the four SSc signature modules (two IFN [M1.2 and M3.4], neutrophil [M 5.15], and NK/cytotoxic [M3.6] and the expected longitudinal course of mRSS and FVC% up to 72 months post-randomization.

In the HSCT arm, higher baseline levels for the IFN transcript modules (M1.2 and M3.4) were associated with more rapid declines in mRSS, while no notable findings were seen in the CYC arm (Supplementary Table 3, Supplementary Figure 3). None of the baseline transcript modules impacted the course of FVC% in either the HSCT or CYC arm (Supplementary Table 4). Of note, the baseline IFN transcript module scores were not significantly associated with anti-topoisomerase-I antibodies (Supplementary Table 5).

DISCUSSION:

In the present study, using genome-wide PBC gene expression analysis from SCOT participants 38- and 54-months post-treatment, we show HSCT can normalize the SSc immune dysregulation on a long-term basis. These findings parallel the observed long-term clinical benefits of this treatment modality. CYC treatment for one year (between the baseline- 12-month visits) did not have any long-term effects on the PBC gene expression profile 2-3.5 years after completion of the active treatment period and participants in this treatment arm continued showing a disease-related gene expression signature at the 54-month visit.

In our previous study, HSCT normalized the SSc signature at the 26-month time point (4). The current report extends these analyses to the 38- and 54-month time points. Reflective of the findings at 26 months post-HSCT, the IFN modules were decreased, while the cytotoxic/NK cell signature was increased at 38 months. Furthermore, at 38 months HSCT-treated individuals developed a unique PBC signature reflective of a naïve immune system composed of activated, young B cells (i.e., plasmablasts). This immune composition mirrors neither their baseline status nor healthy controls with a more mature immune system. These results indicate that at the 38-month time point, immune reconstitution is still actively occurring in the HSCT arm. This finding is also consistent with previous reports that HSCT switches the B-cell profile of individuals with SSc towards naïve B cells; however, follow-up time in previous studies did not extend beyond 16 months (10,11). A recent study from the SCOT trial, examining the longitudinal peripheral blood immunoglobulin heavy chain repertoires of study participants, showed that HSCT led to an increase in IgM isotype antibodies bearing a low mutation at 38-month visit which parallels the observed upregulation of plasmablasts and B cells modules at this time point in the current study (12).

A notable finding of this study is the PBC status of HSCT-treated participants 54 months post-treatment. Although the 54 months sample results in part parallel the findings at the 38-month visit showing that HSCT, contrary to CYC, has a long-term normalizing effect on the SSc signature, the HSCT 54-month samples are more similar to healthy controls than HSCT 38-month samples. Specifically, HSCT 54-month samples lack an upregulation of B-cell modules and cell cycle modules observed at the previous time points, indicating that immune reconstitution at the disease transcript level appears complete at 54 month but not at 36 month. Although the sample size of participants in the HSCT arm decreased from 19 to 16 from the 38-month to 54-month visit, it is unlikely that this slight decrease in the sample size is the primary reason for not detecting any differential modules between 54-month samples and healthy controls. These findings indicate that HSCT-treated participants with severe SSc have a healthy, fully immune reconstituted PBC immune profile 4.5 years after treatment. The normalization of the SSc signature, including a decrease in the IFN module, parallels the clinical benefits of HSCT treatment at this time point and provides support for these disease-related pathways as therapeutic targets. However, we cannot directly differentiate the primary HSCT treatment effect from the potential secondary effects (i.e., diminished immune response due to improved disease damage such as decreasing skin and lung fibrosis) at the gene expression level in the present study.

Consistent with previous studies, treatment with CYC during the first year of the trial did not lead to long-term changes in PBC gene expression profile, paralleling published data on lack of long-term clinical benefit of short-term immunosuppressive treatment with CYC (2,3,13).

Although our study provides important data on normalization of SSc PBC gene expression signatures 4.5 years after treatment, future studies are required to determine whether HSCT-treated individuals will maintain a healthy PBC profile past 54 months, including the absence of SSc signatures. Another limitation of the present study is that the longitudinal PBC gene profile could only be examined in the participants with an available biospecimen during the follow-up visit; participants who died or were discontinued once they met another component of event-free-survival criteria (respiratory failure, renal failure, or cardiac failure) prior to the 38-month visit were not included in the present study, therefore, we cannot exclude that our study results are influenced by survival bias and are skewed towards those participants with more favorable clinical outcome.

In summary, the present study provides new knowledge about the long-term effects of HSCT on normalization of SSc signature and present further evidence for the consideration of HSCT as an effective, disease-modifying treatment option for severe SSc.