ATP-Dependent Chromatin Remodeler CHD9 Controls the Proliferation of Embryonic Stem Cells in a Cell Culture Condition-Dependent Manner

Simple Summary Chromodomain-helicase-DNA-binding protein 9 (CHD9) has been implicated in the regulation of gene expression, yet its precise role in the maintenance of mammalian embryonic stem cell (ESC) remains unclear. In the present study, we demonstrated that mouse CHD9 controls the cell cycle of ESCs in a cell culture condition-dependent manner by modulating the accessibility of transcription factors to their target genomic elements. Our study, therefore, has not only established how CHD9 finetunes chromatin structure during animal development but provided a potential target for genetic screening of aberrant development in in vitro produced embryos. Abstract Emerging evidence suggests that chromodomain-helicase-DNA-binding (CHD) proteins are involved in stem cell maintenance and differentiation via the coordination of chromatin structure and gene expression. However, the molecular function of some CHD proteins in stem cell regulation is still poorly understood. Herein, we show that Chd9 knockdown (KD) in mouse embryonic stem cells (ESCs) cultured in normal serum media, not in 2i-leukemia inhibitory factor (LIF) media, causes rapid cell proliferation. This is caused by transcriptional regulation related to the cell cycle and the response to growth factors. Our analysis showed that, unlike the serum cultured-Chd9 KD ESCs, the 2i-LIF-cultured-Chd9 KO ESCs displayed elevated levels of critical G1 phase regulators such as p21 and p27. Consistently, the DNA binding sites of CHD9 overlap with some transcription factor DNA motifs that are associated with genes regulating the cell cycle and growth pathways. These transcription factors include the cycle gene homology region (CHR), Arid5a, and LIN54. Collectively, our results provide new insights into CHD9-mediated gene transcription for controlling the cell cycle of ESCs.


Introduction
Chromodomain-helicase-DNA-binding (CHD) proteins, a subfamily member of the ATP-dependent chromatin-remodeling families, are largely evolutionary conserved. Eukaryotic CHDs have been implicated in diverse developmental processes including zygotic reprogramming, gastrulation and organogenesis [1][2][3]. CHD proteins (CHD1-9) are categorized into three subfamilies (subfamily I-III) based on their domain structures and sequence similarities, and their critical roles in stem cell function have been elucidated. Subfamily I proteins (CHD1 and CHD2) contain dual chromodomains

CHD9 is Dispensable for the Expression of Pluripotency Markers
To examine the expression pattern of Chd genes during mouse ESC differentiation, a publicly available RNA-Seq dataset (GSE114219) was obtained and re-analyzed to compare the fragments per kilobase of transcripts per million mapped reads (FPKM) values of Chd1~9 (Figure 1a). Chd4 expression was the most abundant, whereas Chd5 expression was the least abundant during ESC differentiation. Among the subfamily III Chd genes, Chd6 and 7 were induced immediately after differentiation and their expression levels were maintained at least up to D6 of differentiation. In contrast, a mild increment in Chd8 and a robust increment in Chd9 were detected at D2, though their levels decreased sharply at D4 and D6. Next, we sought to determine the role of CHD9 in the maintenance of ESC pluripotency using the short hairpin (sh) RNA KD approach. To this end, two shChd9 were designed, tested for their KD efficiency, and found to repress~80% of endogenous Chd9 expression (Figure 1b). Immunofluorescence, as well as AP staining, was performed to examine the levels of ESC markers, including OCT4, SOX2, NANOG, and SSEA-1. No gross changes in the levels of these markers were detected upon Chd9 KD in the ESCs (Figure 1c,d). Moreover, the transcript levels of the pluripotent genes in the Chd9 KD ESCs were indistinguishable from those in the control ESCs ( Figure 1e). These data implied that CHD9 is dispensable for the expression of pluripotent markers. To examine the expression pattern of Chd genes during mouse ESC differentiation, a publicly available RNA-Seq dataset (GSE114219) was obtained and re-analyzed to compare the fragments per kilobase of transcripts per million mapped reads (FPKM) values of Chd1~9 (Figure 1a). Chd4 expression was the most abundant, whereas Chd5 expression was the least abundant during ESC differentiation. Among the subfamily III Chd genes, Chd6 and 7 were induced immediately after differentiation and their expression levels were maintained at least up to D6 of differentiation. In contrast, a mild increment in Chd8 and a robust increment in Chd9 were detected at D2, though their levels decreased sharply at D4 and D6. Next, we sought to determine the role of CHD9 in the maintenance of ESC pluripotency using the short hairpin (sh) RNA KD approach. To this end, two shChd9 were designed, tested for their KD efficiency, and found to repress ~80% of endogenous Chd9 expression ( Figure 1b). Immunofluorescence, as well as AP staining, was performed to examine the levels of ESC markers, including OCT4, SOX2, NANOG, and SSEA-1. No gross changes in the levels of these markers were detected upon Chd9 KD in the ESCs (Figure 1c and d). Moreover, the transcript levels of the pluripotent genes in the Chd9 KD ESCs were indistinguishable from those in the control ESCs ( Figure 1e). These data implied that CHD9 is dispensable for the expression of pluripotent markers.  SSEA-1(IF)) was determined via immunofluorescence staining (d) and qRT-PCR (e), respectively. There was no significant difference in the transcription of Nanog, Oct3/4, and Sox2 upon Chd9 KD. The ESCs were immunostained with anti-NANOG, OCT3/4, SOX2, and SSEA-1 antibodies; scale bars in the merged images of low magnification in the left panels are 200 µm and of high magnification in the right panels are 100 µm.

Chd9 Silencing Promotes ESC Proliferation
Intriguingly, we found that the Chd9-depleted ESCs exhibited rapid cell proliferation. To precisely determine CHD9 function in ESC proliferation, cell growth was assessed up to 7 d after 1000 ESCs were seeded in normal serum media ("serum-ESCs") or 2i-LIF media ("2i-ESCs"). As shown in Figure 2a, the Chd9 KD serum-ESCs grew around two-times faster than the control serum-ESCs, whereas the Chd9 KD 2i-ESCs grew at a normal rate. To confirm these findings, the Chd9 KD ESCs were subjected to flow cytometric analysis. As shown in Figure 2b, the KD serum-ESCs exhibited a modestly shortened G1 phase compared to the control serum-ESCs, whereas the KD 2i-ESCs exhibited a significantly extended G1 phase compared to the control 2i-ESCs.

Chd9 Silencing Promotes ESC Proliferation
Intriguingly, we found that the Chd9-depleted ESCs exhibited rapid cell proliferation. To precisely determine CHD9 function in ESC proliferation, cell growth was assessed up to 7 d after 1000 ESCs were seeded in normal serum media ("serum-ESCs") or 2i-LIF media ("2i-ESCs"). As shown in Figure 2a, the Chd9 KD serum-ESCs grew around two-times faster than the control serum-ESCs, whereas the Chd9 KD 2i-ESCs grew at a normal rate. To confirm these findings, the Chd9 KD ESCs were subjected to flow cytometric analysis. As shown in Figure 2b, the KD serum-ESCs exhibited a modestly shortened G1 phase compared to the control serum-ESCs, whereas the KD 2i-ESCs exhibited a significantly extended G1 phase compared to the control 2i-ESCs.

Chd9 Silencing Alters Gene Expression Related to the Cell Cycle
To unravel the mechanism by which Chd9 KD results in different patterns of cell proliferation depending on the culture conditions, publicly available data of a microarray (GSE64819) performed in Chd9 KD serum-ESCs and RNA-Seq (GSE148803) performed in Chd9 KO 2i-ESCs were re-analyzed. In the analysis of GSE64819, 569 downregulated DEGs and 527 upregulated DEGs were identified in the Chd9 KD serum-ESCs. In contrast, 488 downregulated DEGs and 150 upregulated DEGs were identified in the Chd9 KO 2i-ESCs (Figure 3a). Given that the upregulation of both p21 (encoded by Cdkn1a) and p27 (encoded by Cdkn1b) is required for the extension of the G1 phase in the 2i-ESCs, we examined the expression of these genes. As shown in Figure 3b, both genes were enhanced in the Chd9 KO 2i-ESCs, whereas both genes were not changed in the Chd9 KD serum-ESCs. A biological processes network analysis of upregulated DEGs showed that, among the various terms, genes related to "regulation of cell population proliferation" and "response to growth factor" were altered

Chd9 Silencing Alters Gene Expression Related to the Cell Cycle
To unravel the mechanism by which Chd9 KD results in different patterns of cell proliferation depending on the culture conditions, publicly available data of a microarray (GSE64819) performed in Chd9 KD serum-ESCs and RNA-Seq (GSE148803) performed in Chd9 KO 2i-ESCs were re-analyzed. In the analysis of GSE64819, 569 downregulated DEGs and 527 upregulated DEGs were identified in the Chd9 KD serum-ESCs. In contrast, 488 downregulated DEGs and 150 upregulated DEGs were identified in the Chd9 KO 2i-ESCs ( Figure 3a). Given that the upregulation of both p21 (encoded by Cdkn1a) and p27 (encoded by Cdkn1b) is required for the extension of the G1 phase in the 2i-ESCs, we examined the expression of these genes. As shown in Figure 3b, both genes were enhanced in the Chd9 KO 2i-ESCs, whereas both genes were not changed in the Chd9 KD serum-ESCs. A biological processes network analysis of upregulated DEGs showed that, among the various terms, genes related to "regulation of cell population proliferation" and "response to growth factor" were altered in the Chd9 KD serum-ESCs ( Figure 3c). However, these BPs were not identified in the Chd9 KO 2i-ESCs. Although a batch effect was observed in the KD experiment, in general, most of the genes related to "regulation of cell population proliferation" were enhanced in the Chd9 KD serum-ESCs ( Figure 3d). ESCs. Although a batch effect was observed in the KD experiment, in general, most of the genes related to "regulation of cell population proliferation" were enhanced in the Chd9 KD serum-ESCs ( Figure 3d).

CHD9 Binding Regions Overlaps with the Transcription Factor Binding Sites Associated with the Regulation of the Cell Cycle
Next, to determine the CHD9 binding sites, publicly available data of a Chd9 ChIP-Seq (GSE64825) carried out in mouse ESCs was re-analyzed. Our analysis identified 2614 genomic regions that Chd9 directly binds to. Further downstream analyses, including gene ontology biological process (GOBP) and KEGG pathway, were performed. As shown in Figure 4a, the "cell cycle" (161 genes)related GOBP was highly enriched. "Cell division" (109 genes) was also identified among the top 10 enriched GOBP in the analysis. Other GOBPs associated with transcription and translation were shown to be altered. The KEGG pathway analysis also identified gene expression-and cell cyclerelated terms such as "ribosome" (71 genes) and "cell cycle" (38 genes). Next, we performed a DNA motif analysis to identify genomic regions that CHD9 binds to (Figure 4b). Then, the identified CHD9 binding motifs were examined to investigate whether transcription factor (TF) binding DNA sequences could be found in the motifs. The analysis showed that some CHD9 target sequences overlapped with known TFs that control the cell cycle. The predicted TFs included cell cycle gene homology region (CHR), AT-rich interaction domain 5A (Arid5a), and LIN54. In particular, the cell cycle-related CHR binding sequence, TTTAAA, was found in the cyclin B1 (Ccnb1) promoter. CCNB1

CHD9 Binding Regions Overlaps with the Transcription Factor Binding Sites Associated with the Regulation of the Cell Cycle
Next, to determine the CHD9 binding sites, publicly available data of a Chd9 ChIP-Seq (GSE64825) carried out in mouse ESCs was re-analyzed. Our analysis identified 2614 genomic regions that Chd9 directly binds to. Further downstream analyses, including gene ontology biological process (GOBP) and KEGG pathway, were performed. As shown in Figure 4a, the "cell cycle" (161 genes)-related GOBP was highly enriched. "Cell division" (109 genes) was also identified among the top 10 enriched GOBP in the analysis. Other GOBPs associated with transcription and translation were shown to be altered. The KEGG pathway analysis also identified gene expression-and cell cycle-related terms such as "ribosome" (71 genes) and "cell cycle" (38 genes). Next, we performed a DNA motif analysis to identify genomic regions that CHD9 binds to (Figure 4b). Then, the identified CHD9 binding motifs were examined to investigate whether transcription factor (TF) binding DNA sequences could be found in the motifs. The analysis showed that some CHD9 target sequences overlapped with known TFs that control the cell cycle. The predicted TFs included cell cycle gene homology region (CHR), AT-rich interaction domain 5A (Arid5a), and LIN54. In particular, the cell cycle-related CHR binding sequence, TTTAAA, was found in the cyclin B1 (Ccnb1) promoter. CCNB1 is essential for the regulation of the cell cycle; it interacts with other proteins such as the dimerization partner (DP), RB-like, E2F4, and MuvB (DREAM) complex [28]. Next, we sought to identify direct CHD9 target genes by analyzing those that changed their expression level upon Chd9 silencing. Thirty-three CHD9 target genes repressed by Chd9 KD and 55 enhanced by Chd9 KD were identified in the comparison between the microarray and ChIP-Seq data. Our GOBP analysis showed that genes associated with chromatin modification such as "nucleosome assembly," "chromatin silencing," and "regulation of transcription" were significantly affected (Figure 4c). These results suggest that the disruption of CHD9 results in changes in the chromatin structure at regions related to cell growth and causes aberrant proliferation. is essential for the regulation of the cell cycle; it interacts with other proteins such as the dimerization partner (DP), RB-like, E2F4, and MuvB (DREAM) complex [28]. Next, we sought to identify direct CHD9 target genes by analyzing those that changed their expression level upon Chd9 silencing. Thirty-three CHD9 target genes repressed by Chd9 KD and 55 enhanced by Chd9 KD were identified in the comparison between the microarray and ChIP-Seq data. Our GOBP analysis showed that genes associated with chromatin modification such as "nucleosome assembly," "chromatin silencing," and "regulation of transcription" were significantly affected (Figure 4c). These results suggest that the disruption of CHD9 results in changes in the chromatin structure at regions related to cell growth and causes aberrant proliferation.

ATP-dependent chromatin remodelers have been shown to be implicated in the regulation of stem cell maintenance and differentiation, yet the biological function of CHD9 in ESCs is still unknown.
In the present study, we demonstrated that CHD9 is required for the precise regulation of the cell cycle in mouse ESCs. Upon Chd9 silencing, the serum-ESCs exhibited a mildly shorter G1 phase as well as a higher proliferation rate compared to the control ESCs. However, the extension of the G1 phase cannot be seen in the 2i-ESCs. The chromatin structure in ESCs is highly dynamic throughout the cell cycle. It is coordinated by stem cell factors such as OCT3/4 and SOX2 and closely related to the transcriptional regulation of genes that specify lineage commitment or maintain the self-renewal capacity. For example, the loss of OCT3/4 at the time of M-G1 transition leads to failure in ESC maintenance and an increased tendency of differentiation to neuroectoderm and mesendoderm [29]. This is most likely due to the increased chromatin accessibility near the differentiation genes in the absence of OCT3/4; it was observed even when OCT3/4 was degraded for a short time window during the M-to-G1 transition. This suggests that the coordinated regulation in the G1 phase by OCT3/4 and other TFs is critical for maintaining chromatin structure and pluripotency [30]. Consistent with published literature, Chd9 KD altered the expression of genes involved in the cell cycle and responded to growth factors. In particular, our analysis showed that the expression of p21 and p27 was not altered in the Chd9 KD serum-ESCs, whereas the levels of both p21 and p27 were enhanced in the Chd9 KO 2i-ESCs. The data explain why only the Chd9 KD serum-ESCs exhibited a truncated G1 phase and rapid proliferation. It has been shown that 2i-ESCs express high levels of p16, p21, and p27, and the depletion of either p21 or p27 independently leads to no change in the length of the G1 phase. The study demonstrated that both p21 and p27 are required for the extended G1 phase in the 2i-ESCs [26]. A recent report demonstrated that mouse CHD9 is dispensable for embryogenesis and postnatal development; however, it is required for gene expression [18]. Our analysis also showed that Chd9 depletion changes gene expression without altering ESC maintenance and the expression of stem cell markers. Ooga et al. also showed that mouse CHD9 is necessary for growing oocytes to allow the loosening of the chromatin structure during the acquisition of pluripotent potential after fertilization [31]. Although our study has revealed a crucial role of Chd9 in the control of cell cycle depending upon ESC culture conditions, we cannot completely rule out the possibility that the identified Chd9 mechanism is due to the meta-analysis in the differentially produced cohorts (KD ESCs vs. KO ESCs) or different targeting sequences for Chd9 ablation. However, given that the direct comparison between KO and KD was only made in the determination of p21 and p27 levels, our bioinformatic analysis in Chd9 KD serum-ESCs is still worthwhile to explain mechanistic aspects of Chd9 functions. Furthermore, our study clearly demonstrates that Chd9 deficiency gives rise to a distinct role in the regulation of the ESC cell cycle at the cellular level.
Besides the regulatory TFs, the epigenetic landscape in the 2i-ESCs differs from that in the serum-ESCs. For example, the DNA methylation level is globally reduced in the 2i-ESCs compared to that in the serum-ESCs. Furthermore, histone H3K9me3 is enriched at the hypermethylated genomic regions, including endogenous retroviruses (ERVs) and imprinting genes [32]. Recently, Asenjo et al. demonstrated that the recruitment of Polycomb repressive complex 2 (PRC2) subunits to genes involved in the specification of lineages is tightly regulated across the cell cycle in mouse ESCs. EZH2, a catalytic subunit of the PRC2 complex, and JARID2, a protein enhancing PRC2 activity are enriched at bivalent gene promoters during the S and G2 phases, whereas EPOP, a PRC2 subunit activating bivalent genes, is enriched during the G1 phase [33]. Studies suggest that ATP-dependent chromatin remodelers play an important role in cell cycle control. Human INO80 was shown to negatively regulate p21 expression via p53-mediated binding to the p21 promoter [34]. INO80 KD led to extended transition from G2/M to G1. The mammalian SWI/SNF complex (also known as the BAF complex) regulates the cell cycle depending on the composition of the complex subunits [35]. ARID1B (a subunit of the SWI/SNF complex)-depleted cells exhibited delayed DNA synthesis, whereas ARID1A (an alternative subunit of ARID1B in the SWI/SNF complex)-depleted cells led to a slight acceleration of the cell cycle [35]. The context-dependent subunit switch in the complex was shown to be important for the regulation of the expression of cell cycle regulators, including E2F genes [36]. Furthermore, BRG1 (a catalytic subunit of the SWI/SNF complex) regulates the proliferation of hepatocytes by modulating cell cycle-related genes such as Ccnb1, Ccnb2, Cdk1, and Cdc20 [36]. Another context-dependent-Chd9 function in the regulation of the cell cycle was also demonstrated in a recent study. Xia et al., showed that CircPDZD8, a circular RNA, elevates CHD9 expression by interfering with miR-197-5p in gastric cancer, and-interestingly-knocking down CHD9 in gastric cancer cells decreases cell growth [37]. Additionally, some CHD proteins play roles in the regulation of the cell cycle and survival. CHD1 was shown to be essential for the maintenance of open chromatin and pluripotency in ESCs, and its silencing causes the aberrant regulation of pluripotency [5]. Chd7 KD causes the inhibition of cell proliferation and neurogenesis in vitro by repressing Hes5 and N-myc expression, whereas Chd7 overexpression leads to the acceleration of cell proliferation in human ESC-derived neural stem/progenitor cells [38]. CHD7 is also critical for the proliferation of neural stem cells; its loss causes the activation of p53, resulting in human CHARGE syndrome [38,39]. Chd8 mutation leads to induced apoptosis during early embryo development in mice by the activation of p53 [40]. CHD8 represses neuronal pathway-related transcripts, affecting brain development and synapse formation from induced pluripotent stem cells [41]. It is generally accepted that the synthesis of histone proteins is active during the S phase, such that nucleosome assembly quickly occurs in parallel with DNA replication [42]. Liu et al. also showed that the enrichment of histone H3K27 acetylation (H3K27ac) at proximal and distal enhancers ensures the maintenance of stem cell identity and the cell cycle during mitosis [43].
Our de novo motif analysis in CHD9 ChIP-Seq data showed that the CHD9 protein binds to genomic regions that overlap with cell cycle-related TFs binding DNA motifs, including CHR, LIN54, and Arid5a. It has been shown that CHR is present at the Ccnb1 promoter and that p53-p21-DREAM functions through CHR to regulate the expression of some critical cell cycle regulators, including CCNA2, CCNB1, and POLD1 [44,45]. Moreover, CHR was shown to modulate the cell cycle or programmed cell death in humans and mice [46][47][48]. The DNA binding protein LIN54 also interacts with CHR elements in the cell cycle-associated promoter [48]. Arid5a regulates the quiescence stage by increasing the G0/G1 phase in human cancer cells; thus, it is regarded as a quiescence-associated marker [46]. Therefore, the ChIP-Seq analysis reveals how CHD9 deficiency leads to an enhancement of cell proliferation in the serum-ESCs.

Alkaline Phosphatase (AP) Staining and Immunofluorescence
For immunofluorescence staining, the cells were fixed with 4% paraformaldehyde, blocked using blocking solution (3% BSA and 2% serum of host secondary antibodies with 0.

RNA-Seq, Microarray, and ChIP-Seq Analyses
RNA-Seq data performed in mouse embryoid bodies (GSE114219) and Chd9 knockout ESCs (GSE148803) were obtained from Gene Expression Omnibus (GEO). The quality of the sequencing reads of the data was controlled using FastQC [49], and the reads were aligned to UCSC mm10 genome assembly using the following STAR (v2.7.1a, NY, USA) [50] options: clip3pNbases argument set as 1. Cuffquant and Cuffnorm of the Cufflinks toolset (version 2.2.1, Seattle, WA, USA) [51] were used for the quantification of the normalized expression of each gene.
Microarray data performed in Chd9 KD ESCs (GSE64819) were obtained from GEO. The microarray data were normalized and fitted into a linear model using the LIMMA (v3.40.6, QLD, Austria) tool [52] in R package (version 3.6.2, Vienna, Austria) [53]. Gene sets involved in the Gene ontology (GO) regulation of cell population proliferation and fold change of gene expression were obtained from Mouse Genome Informatics [54] and the LIMMA tool [55], respectively. The resultant heatmap showing the expression of the genes was generated using the heatmap2 function of the gplot (v3.0.4, Tel Aviv, Israel) [56] package and R tool. Differentially expressed genes (DEGs) were determined by comparison between the Chd9 KD and control ESCs using LIMMA's fold change and p-value data. With the list of DEGs, GO analysis was performed using the DAVID tool (version 6.8, Frederick, MD, USA) [57]. ClueGo (v2.5.1, Cordeliers Research Center, France) [58] plugged-in the Cytoscape (version 3.7.2, La Jolla, CA, USA) tool [59] was used to visualize the correlation of GO (biological pathway, BP) terms and KEGG pathways.
A publicly available CHD9 ChIP-Seq dataset (GSE64825) was downloaded from GEO. The read quality of ChIP-Seq was assessed using FastQC. Bowtie2 (v2.3.4.1, Maryland, USA) [60] was used to map the reads to UCSC mm9 mouse genome assembly using the following option: the trim3 1 option. The resultant files were used to call peaks in the MACS2 tool (v2.1.0, MA, USA) [61]. The peak regions calculated by MACS2 were used to find CHD9 binding motifs using the HOMER tool (v4.10.3, La Jolla, CA, USA) [62]. The MACS2 output bedgraph files were used as input for seqMINER (v1.3.4, CU de Strasbourg, France) [54] in which genomic regions surrounding all of the mm9 transcription start sites were visualized. Six gene clusters were calculated via the unsupervised clustering method. Genes in each cluster having signals around TSS were selected for comparison with DEGs from the microarray data. The commonly overlapping genes were used for GOBP analysis using DAVID.

Statistical Analysis
All the bar graph data were generated using Prism 5.0 (GraphPad Software, San Diego, CA, USA). Normality tests and equal variance tests were performed using SigmaPlot 12.0 (Systat Software Inc. Erkrath, German). If the data failed the normality test, a rank-sum test was used and plotted in a scatter plot. One-way ANOVA and two-way ANOVA with a Holm-Sidak test as post hoc analysis were used for the qRT-PCR data of the ectoderm marker and proliferation data, respectively. Unpaired t-test analysis was performed on the rest of the data. The error bars in the data represent the standard error of the mean (SEM). Statistical significance was determined at a p-value lower than 0.05(*) or 0.001(***).

Conclusions
In summary, our study demonstrates that CHD9 regulates ESC proliferation by regulating the expression of cell cycle-related genes in a cell culture condition-dependent manner. In addition, our analysis suggests that CHD9 directly modulates the accessibility of cell cycle-related TFs to their target genes in the serum-ESCs. The analysis of the exact mechanism and partner proteins of the CHD9 machinery, which regulates cell cycle-related genes, is warranted in follow-up studies.