Allelic Losses at 1p36 and 19q13 in Gliomas: Correlation with Histologic Classification, Definition of a 150-kb Minimal Deleted Region on 1p36, and Evaluation of CAMTA1 as a Candidate Tumor Suppressor Gene

Diffuse gliomas are the most common primary tumors of the central nervous system. The major subtypes include astrocytoma, oligodendrogliomas, and so-called “oligoastrocytoma.” Prior studies have shown that ∼70% to 85% of oligodendrogliomas and 20% to 30% of astrocytomas have allelic losses on chromosome arm 1p suggesting the presence of a tumor suppressor gene (1–4). However, the latter remains elusive because most deletions are very large, often involving the entire 1p arm.

Loss of genetic material on 1p is not unique to glioma and occurs in many human cancers. In a study of 683 solid tumors of various types (5), the prevalence of loss of heterozygosity (LOH) on 1p ranged from 30% to 64% depending on tumor site. The most frequent 1p LOH was seen in breast, lung, endometrial, ovarian, and colorectal carcinomas. High percentages of LOH were identified at 1p36.3, 1p36.1, 1p35-34.3, 1p32, and 1p31 regions, with tumor-specific patterns. The most extensive 1p deletion mapping in search for a tumor suppressor has been done in neuroblastomas, known to have 1p losses (predominantly at 1p36) in ∼30% of cases. Genes on 1p previously subjected to mutation analysis as candidate tumor suppressors for neuroblastoma or glioma have included TP73, RAD54L, KIF1B, HKR3, UBE4B/UFD2, EXTL1, and CHD5 (6–17). However, no strong candidate for the 1p tumor suppressor gene in either neuroblastomas or gliomas has thus far been identified. The putative 19q tumor suppressor also remains elusive, although the p190RhoGAP gene has recently been proposed as a candidate (18, 19).

Remarkably, in spite of neither tumor suppressor gene being clearly identified, the detection of combined allelic losses at 1p and 19q is already in widespread clinical use in neuro-oncology and is becoming part of the standard of care for patients with oligodendrogliomas because of its strong correlation with histology and chemotherapy response (1, 2). As part of a prospective analysis of 205 gliomas, we identified 11 tumors with small losses at 1p36 that allowed us to substantially narrow the commonly deleted region, thereby pinpointing CAMTA1 (20) as a new candidate tumor suppressor gene. We have also sought to better define the histologic correlates of different patterns of 1p and 19q loss (combined versus isolated and extensive versus segmental) in the subset 110 gliomas studied for both loci.

Specimens. Among 205 consecutive gliomas studied for 1p LOH at Memorial Sloan-Kettering Cancer Center (January 2000-September 2003), 112 tumors were evaluated for combined 1p/19q loss (19q LOH analysis was done prospectively in all cases evaluated starting in January 2003, as well as retrospectively based on DNA availability). These 112 gliomas included low-grade tumors (24 oligodendrogliomas, 18 diffuse astrocytomas, and 15 oligoastrocytomas) and high-grade tumors (18 anaplastic oligodendrogliomas, 17 anaplastic astrocytomas, and 18 anaplastic oligoastrocytomas) as well as two other brain tumors with oligodendroglial components. Tumor samples consisted of routinely processed formalin-fixed, paraffin- embedded tissue. For each tumor, histologic subtype and grade were documented by a neuropathologist (M.K.R.). An H&E-stained section was obtained before DNA extraction to evaluate the proportion of tumor in paraffin block. If tumor content was <90%, areas containing pure tumor were dissected. Matched constitutional DNA was obtained from peripheral blood or nail clippings. Frozen tumor material, obtained in a limited number of the above cases under a protocol approved by the Memorial Sloan-Kettering Cancer Center Institutional Review Board, was used for RNA extraction for quantitative real-time reverse transcriptase-PCR (RT-PCR).

DNA Extraction. Tumor DNA was extracted from paraffin-embedded tissue using the DNeasy Tissue Kit (QIAGEN, Inc., Valencia, CA) according to the manufacturer's protocol. Following DNA extraction, all tumor samples were subjected to control gene (PGK) amplification to assess DNA integrity. Constitutional DNA from peripheral blood leukocytes was isolated using the standard phenol/chlorophorm extraction method with ethanol precipitation. In cases when peripheral blood was unavailable, patient's nail clippings were used as a source of normal DNA. DNA extraction from nails was done using a protocol modified from published methods (21, 22) with reagents from the PUREGENE DNA Purification Kit (Gentra Systems, Inc., Minneapolis, MN ). Briefly, 1 to 5 mg of nail clippings were snap-frozen in liquid nitrogen, finely ground, treated with 150 μL of cell lysis solution, and 1.25 μL of proteinase K and incubated at 55°C for 72 hours. The lysate was treated with 0.75 μL of RNase A solution (4 mg/mL) and incubated for an additional hour. Following incubation, the sample was cooled on ice for at least 5 minutes. Cell lysate was treated with 50 μL of protein precipitation solution, vortexed, and centrifuged at 14,000 × g for 3 minutes. The precipitated protein formed a tight pellet. Supernatant containing DNA was transferred to a clean 1.5-mL tube and treated with 150 μL of 100% isopropanol and 0.5 μL of glycogen solution (20 mg/mL). The sample was gently mixed and centrifuged at 14,000 × g for 5 minutes. Supernatant was discarded and the pellet was allowed to air-dry. DNA pellet was washed with 150 μL of 70% ethanol, centrifuged at 14,000 × g for 1 minute and allowed to completely air-dry. DNA was re-suspended in 30 μL of DNA hydration solution, incubated at 65°C for 1 hour, and allowed to completely dissolve at room temperature overnight. Average DNA yield was 0.5 to 2 μg.

Loss of Heterozygosity Analysis. Constitutional DNA/tumor DNA pairs were evaluated by standard PCR-based LOH assays. All cases were initially studied using at least three microsatellite markers mapped to the distal portion of the chromosome within the region of previously reported minimal 1p losses (tel-D1S468, D1S548, D1S1612, D1S1592, D1S552, D1S496-cent). Segmental loss was defined as evidence of interstitial or small terminal deletions, as manifested by LOH at one or more loci with retention of heterozygosity at the centromeric end of the evaluated region, with or without retention of heterozygosity at the telomeric end. Cases with segmental 1p36 deletions were studied with an expanded panel of 15 markers mapped to p36.32 to p34.3 and spanning ∼32 Mb (tel- D1S171, D1S468, D1S2145, D1S2870, D1S253, D1S2642, D1S214, D1S2694, D1S548, D1S2666, D1S508, D1S1612, D1S1592, D1S552, and D1S496-cent). Information on these loci was obtained from various public databases [Genome Database (http://www.gdb.org), National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov), Ensembl Genome Browser (http://www.ensembl.org/Homo_sapiens/)].

Allelic losses at 19q were evaluated using 3 polymorphic markers mapped to a commonly deleted region at q13.3 (D1S219, D1S412, and PLA4C2G). Oligonucleotide primer pairs were custom synthesized at Proligo LLC (http://www.gensetoligos.com). Approximately 100 ng of DNA and 15 pmol of each primer (forward primer labeled with 6-FAM) were used for each PCR reaction. All PCR products were first run on a 2% agarose gel to check the efficacy of PCR amplification and then analyzed by capillary electrophoresis on 3100 Avant Genetic Analyzer (Applied Biosystems, Foster City, CA). LOH was scored as present when the tumor sample showed reduction in height of one of two peaks by >50%. This was compared with the ratio of peak heights of the two alleles in the corresponding constitutional DNA. The consistency of all significant findings was confirmed by repeat analysis.

Mutation Analysis. The exons encoding conserved regions of CAMTA1 (Ensembl: ENSG00000171735) were evaluated for somatic mutations by PCR amplification followed by direct sequencing. Primers for the amplification of specific exons (listed in Table 1) were selected in the flanking intronic sequences. The GC-rich (78% GC content) region of exon 1 was particularly difficult to amplify. The specific 386-bp product was obtained only with the use of the approach described by Bachmann et al. (23) that included partial substitution of dGTP in the standard PCR mix with 7-Deaza-dGTP [150 μmol/L of 7-Deaza-dGTP (Roche, Indianapolis, IN) + 50 μmol/L dGTP per reaction, instead of 200 μmol/L dGTP]. We also did mutation analysis of selected anonymous expressed sequence tags (EST) within a large intron 5 of CAMTA1 that aligned with a region of consistent LOH within BAC clone RP3-453P22 (Genbank accession no. Z97987; GI: 2769643). All primer sequences and PCR product sizes are listed in Tables 1 and 2.

CAMTA1 Expression Analysis in Normal Tissues. Expression of CAMTA1 in normal human tissues was evaluated by RT-PCR and Northern blotting using Clontech Multiple Tissue cDNA Panel and Human Brain MTN Blot. Gene-specific primers for RT-PCR were exon 2 sense CGTTTCCCAAAGTGTATTCTGCGGA and exon 6 antisense ATGGACATAGCAGCCGTACAAGCA. The identity of RT-PCR product was confirmed by direct sequencing. Northern blotting was done using a [α-³²P] dCTP-labeled probe that represented a PCR-amplified portion of CAMTA1 exon 9 spanning 1 kb (primers, sense GTGTTCATGTCAGAGGTCACCA and antisense AGGAGATGTCAAAGTGGTCCAG). The hybridization was carried out at 63°C in ExpressHyb solution (Clontech, Palo Alto, CA) for 2 hours. The blot was washed according to the manufacturer’s protocol and exposed to a phosphorimaging screen (Bio-Rad, Hercules, CA) for 24 hours.

CAMTA1 Expression in Gliomas by Quantitative Reverse Transcriptase-PCR. Total RNA was extracted from frozen tumors using RNA STAT-60 isolation reagent (Tel-Test, Inc., Friendswood, TX) and was quantified on a spectrophotometer. One microgram of RNA was reverse-transcribed using SuperScript II Reverse Transcriptase (Invitrogen, Carlsbad, CA), and 10 μL of cDNA were used in quantitative RT-PCR reaction. Quantitative real-time RT-PCR was done on an iCycler (Bio-Rad) using a commercially available CAMTA1 primers/Taqman probe set (Applied Biosystems) spanning exon 10 to 11 boundary. The thermal cycling conditions comprised an initial denaturation step at 95°C for 10 minutes and 50 cycles at 95°C for 30 seconds and 60°C for 1 minute. Each sample was normalized using TATA box-binding protein transcripts as an endogenous housekeeping gene control. Normal adult brain total RNA (Clontech) served as a reference standard for calculation of relative expression values. Experiments were done in triplicate for each data point. The delta-Ct value of the sample was determined by subtracting the average Ct value of CAMTA1 from the average Ct value of the TBP gene. The delta-delta-Ct value was then calculated by subtracting the delta-Ct value of the sample from that of the normal adult brain RNA sample from the same batch. The CAMTA1 expression level relative to normal brain was then estimated as 2^{−(delta-delta-Ct)}. Differences between the CAMTA1 expression levels in different subsets of gliomas were evaluated by the Mann-Whitney test.

Strong Association of Combined 1p/19q Loss of Heterozygosity with Classic Oligodendroglioma Morphology. The results of 1p/19q LOH analysis are summarized in Table 3. We found no significant difference in 1p/19q status between low-grade and anaplastic oligodendrogliomas (1p LOH, 83% versus 83%; 19q LOH, 66% versus 78%; combined 1p/19q loss, 66% versus 72%). Because WHO grade showed no significant impact on the proportions of patterns of 1p and 19q losses in the other two groups (astrocytomas and oligoastrocytomas), these data were combined for each histologic category in the remaining analyses. Thus, 1p and 19q losses, either separately or combined, were significantly more common in oligodendrogliomas than in either astrocytomas or oligoastrocytomas (P < 0.0001). No astrocytic tumors and only 6 of 30 oligoastrocytomas had combined 1p/19q loss.

Low Frequency of Segmental 1p Losses and Association with Astrocytic Histology. Among 205 consecutive gliomas analyzed for 1p LOH, we found only 11 tumors (∼5%) with segmental 1p deletions. However, this may be an underestimate because small deletions can only be excluded in cases where all loci are informative. The group of 11 tumors with segmental 1p losses consisted of six neoplasms exhibiting either pure oligodendroglioma morphology or having oligodendroglioma component [3 pure oligodendrogliomas, 1 oligoastrocytoma, and 2 tumors of other histologies (glioblastoma and a low-grade glioneuronal neoplasm of unconventional type)] and five pure astrocytomas. The segmental losses were interstitial in eight cases and small terminal in three cases, based on the informative loci available (Fig. 1). Small segmental 1p losses showed a strikingly uneven distribution between classic oligodendroglioma and astrocytic groups. Segmental deletions were significantly more common in astrocytomas than in classic oligodendrogliomas (P = 0.006; Table 3). Segmental 1p losses in oligodendroglioma tumors made up only 9% of 1p-deleted cases, consistent with the 5% to 10% range found in other studies (2, 24–26). In contrast, in astrocytic tumors 1p losses are much less common but when present are often segmental (5 of 6, 83%).

Definition of Minimal Region of 1p Loss Contained within CAMTA1. Initial 1p LOH analysis of tumors with partial deletions showed consistent allelic loss present in oligodendroglial and some astrocytic tumors in the telomeric region of 1p distal to D1S1612 (p36.23, 7.8 Mb). This region was further analyzed by a detailed LOH mapping using 15 microsatellite markers as shown in Fig. 1. All informative tumors in the oligodendroglial group and two of three informative astrocytomas showed allelic loss at D1S548 (p36.23, 7.1 Mb). Results in four cases are shown in Fig. 2. The minimal deleted region (MDR) was centered on D1S548 between D1S2694 (p36.23, 7.05 Mb) and D1S2666 (p36.23, 7.16 Mb). Remarkably, this MDR measures only 100 to 150 kb and it is the full extent of deletion in one tumor (05**). The latter tumor was classified as a recurrent low-grade glioneuronal neoplasm with predominantly oligodendroglioma morphology. Histologic appearance of oligodendroglial component in tumors 04* and 05** is shown in Fig. 3.

Overall, a similar pattern of segmental 1p allelic loss was present in five of five informative tumors in the oligodendroglial group and two of three informative astrocytomas (Fig. 1). Two astrocytomas showed additional discrete small regions of 1p deletion at D1S171-D1S2694 (tumor AS-08) and D1S1592-D1S496 (tumor AS-09) suggesting either the presence of more than one tumor suppressor genes or a possibility of random allelic losses in astrocytic tumors.

Results of the 19q LOH assays are also shown in Fig. 1. LOH on 19q was detected in two of six (33%) oligodendroglial tumors with segmental 1p loss. This is within the expected range for a group including not only pure oligodendrogliomas but also mixed astrocytic-oligodendroglioma tumors. All four astrocytomas with segmental 1p loss that were informative for the 19q loci tested maintained heterozygosity on 19q.

Lack of Somatic Mutations in CAMTA1. The only gene currently mapped to the D1S2694-D1S2666 MDR is a recently characterized calmodulin-binding transcription activator gene, CAMTA1 (20). We searched for somatic mutations in exons encoding conserved regions of CAMTA1 using a panel of 10 tumors (with matching normal DNA). These 10 tumors included two with small segmental 1p losses (05** and AS-08, Fig. 1), four gliomas (2 oligodendrogliomas, 2 oligoastrocytomas) with well-documented large 1p deletions, and four gliomas (1 oligodendroglioma, 1 astrocytoma, and 2 oligoastrocytomas) without evidence of 1p allelic loss. Cases with deletions were studied based on the classic two-hit model that would require the remaining copy of the candidate tumor suppressor gene to be inactivated by somatic mutation. Cases without deletions were studied based on the alternative hypothesis that inactivation of only one copy of the gene may be pathogenetically significant (and therefore cases without deletions might harbor a somatic mutation instead). The following 15 exons encoding conserved regions of CAMTA1 were sequenced in each case using primers in the flanking intronic sequences: exons 1 to 2 (NLS_BP: putative nuclear localization signal), exons 3 to 7 (CG-1: DNA binding domain), exons 9 to 11 (TIG: DNA binding domain), exons 13 to 15 (ankyrin repeats), and exons 19 to 20 (IQ motif: calmodulin-binding site). Sequencing of these 15 exons showed no somatic mutations in the 10 cases analyzed. As stated above, these 10 cases included 6 cases with 1p36 LOH and 4 cases without 1p36 LOH.

Lack of Somatic Mutations in Expressed Sequence Tags in Intron 5 of CAMTA1. Because the pattern of LOH in one case (05**) suggested a very small interstitial deletion (between D1S2694 and D1S2666) roughly corresponding to intron 5 of CAMTA1, we speculated that this large intron may contain an unidentified gene. Therefore, we analyzed selected ESTs mapped to the intronic region (corresponding to clone RP3-453P22, Genbank accession no. Z97987). The sequences analyzed corresponded to rare ESTs in four small UniGene clusters and one unclustered EST (Table 2). Several single nucleotide polymorphisms were detected, but no somatic mutations were found in the same 10 cases analyzed above.

Expression Pattern of CAMTA1 in Normal Tissues. Among normal human tissues, brain and kidney showed the highest levels of CAMTA1 expression by RT-PCR (Fig. 4). The results are in agreement with publicly available gene expression data (SOURCE http://source.stanford.edu) that show highest expression of CAMTA1 in normal neural tissues. Northern blotting using a probe corresponding to exon 9 of CAMTA1 showed two transcripts (∼8 and 3 kb) present at approximately same levels in samples from the different regions of normal brain (cerebellum, cerebral cortex, medulla, occipital lobe, frontal lobe, temporal lobe, and putamen) and in the spinal cord (Fig. 5). The exon composition of these two CAMTA1 transcripts is under investigation.

Expression of CAMTA1 by Quantitative Reverse Transcriptase-PCR in Gliomas with Defined 1p Status. Frozen tumor material for RNA extraction and quantitative real-time RT-PCR was available in a limited number of gliomas with and without 1p loss. Using undiluted normal adult brain RNA as an arbitrary reference standard, the expression level of CAMTA1 was 18% (median, 11%) that of normal adult brain in nine oligodendroglial tumors with 1p deletion (7 classic oligodendrogliomas, 2 oligoastrocytomas). Expression levels were clustered in a low range (4-26%) in all but one of these nine cases (an oligoastrocytoma that showed a level of 72%). In contrast, the CAMTA1 expression level was 33% (median, 35%) that of normal adult brain in nine oligodendroglial tumors (1 classic oligodendroglioma, 8 oligoastrocytomas) without 1 p deletion, significantly higher than the group with 1p LOH (Mann-Whitney, P = 0.03). In five astrocytomas (all without 1p LOH), the CAMTA1 expression level was 48% that of normal brain (median, 47%), significantly higher than in oligodendroglial tumors with or without 1p LOH (Mann-Whitney, P = 0.02 and P = 0.03, respectively). The results are summarized in Fig. 6.

The association between classic oligodendroglioma morphology and an elevated prevalence of allelic losses on 1p and 19q, either isolated or combined, has been well documented in the literature (2–4). Our data on the frequencies of 1p/19q deletions in different glioma types are in concordance with other studies (2–4). We have confirmed the strong association of 1p/19q LOH and classic oligodendroglioma morphology and shown that for each histologic category, the pattern of 1p/19q losses is similar in WHO grade 2 and grade 3 tumors.

Several partially overlapping 1p MDRs have been previously described in gliomas (2,24–30) and neuroblastomas (31–37). The 1p MDR identified in this study further narrows a known MDR (D1S468-D1S1612) described in gliomas by Smith et al. (2) and overlaps the neuroblastoma 1p36 MDR (ref. 31; Fig. 7).

The compilation of published data on 1p deletion mapping, presented in Fig. 7, suggests that there may be more than one tumor suppressor gene present in the distal portion of 1p. It is possible that p36.3, p36.2, and p34.2-36.1 each contain a tumor suppressor locus, and that the typically large size of 1p deletions reflects selective pressure to inactivate more than one gene. It is therefore also possible that tumors with smaller 1p deletions could be biologically different from those with more extensive 1p deletion. In this study, the subset of tumors with segmental 1p loss showed consistent LOH in a small region of 1p between polymorphic loci D1S2694 and D1S2666, mapped to p36.23. A search of publicly available human genome data reveals only one gene currently mapped to the D1S2694-D1S2666 region, a recently described calmodulin-binding transcription activator, CAMTA1. Although no function has been attributed to any of the CAMTA proteins to date, based on domain organization they likely represent transcription factors. Bouche et al. (20) have shown that CAMTA proteins localize to the nucleus, bind DNA, and can activate transcription. The conserved regions of the CAMTA1 protein include NLS_BP (bipartite nuclear localization signal encoded by exons 1-2 and 5), CG-1 (DNA binding domain thought to provide sequence specificity and encoded by exons 3-7), TIG (nonspecific DNA binding domain encoded by exons 9-11), ankyrin repeats (involved in protein-protein interactions and encoded by exons 13-15), and IQ motif (calmodulin-binding site encoded by exons 19-20). Although expressed in various tissues, the highest expression of CAMTA1 is seen in neural tissues and kidney. The gene is unusually large, spanning ∼1 Mb between 6.5 and 7.5 Mb on 1p and, remarkably, encompasses five polymorphic loci used in the present study (D1S214, D1S2694, D1S548, D1S2666, and D1S508; Fig. 8). Because the D1S2694-D1S2666 MDR is located entirely within intron 5 of CAMTA1, we also examined some ESTs mapped to this intron (Table 2) for somatic mutations but found none.

The predominantly neural expression of CAMTA1 makes it attractive as a candidate 1p36 tumor suppressor in several tumors with neural or neuroectodermal features, including oligodendroglioma, neuroblastoma, and also Ewing's sarcoma where 1p36 LOH has been reported in ∼15% of cases (38, 39). Indeed, the fact that the 1p MDR in gliomas shown here (D1S2694-D1S2666) overlaps the neuroblastoma MDR (D1S2731-D1S2666; (ref. 31), as shown in Fig. 7, is particularly intriguing and suggests that the analysis of CAMTA1 as a candidate tumor suppressor gene should be extended to neuroblastomas.

Mutation analysis of oligodendroglial tumors with 1p LOH showed no evidence of somatic mutations in CAMTA1 exons encoding conserved domains arguing against a classic two hit mechanism. Mutation analysis of gliomas without 1p LOH also showed no evidence of somatic mutations in CAMTA1, suggesting that CAMTA1 mutation does not substitute for 1p deletion in these tumors.

Our finding that CAMTA1 transcript levels in oligodendroglial tumors with 1p LOH were essentially halved compared with cases without 1p LOH supports biallelic expression of CAMTA1. Continued expression from the remaining CAMTA1 allele in 1p-deleted cases is inconsistent with promoter hypermethylation as a second hit but raises the possibility that CAMTA1 haploinsufficiency may be functionally significant. Tumor suppressors for which haploinsufficiency is pathogenetically significant are being described in increasing numbers (for review see ref. 40). Cell cycle regulatory tumor suppressor genes seem especially dosage-sensitive (40). This is intriguing given that CAMTA1 expression levels change during the cell cycle suggestive of involvement in cell cycle regulation, at least in certain cellular contexts (41).

The high prevalence of hemizygous 1p deletions and rarity or absence of homozygous 1p deletions in gliomas and neuroblastomas suggests that there may be a selective advantage for haploinsufficiency of this tumor suppressor gene (possibly CAMTA1), but that loss of the remaining gene copy may be detrimental. Imprinting seems unlikely since there is no evidence for parent-of-origin effects for 1p36 deletions in either gliomas or neuroblastomas (42, 43). Whether the low expression level of CAMTA1 in oligodendroglial tumors without 1p deletion is due to hypermethylation of promoter CpG islands or differences in CAMTA1 promoter activation due to cellular context or merely reflects the presence of other highly expressing cell types in normal adult brain remains to be investigated.

Finally, the findings of two recent studies further strengthen the candidacy of CAMTA1 as an oligodendroglioma tumor suppressor gene. In a glioma gene expression profiling study (44), Mukasa et al. identified CAMTA1 as one of the top differentially overexpressed genes in astrocytomas relative to oligodendroglioma and glioblastomas (or conversely, underexpressed in oligodendroglioma and glioblastoma relative to astrocytomas). These results parallel our quantitative RT-PCR data. The second study is that of Hahn et al. (45), who used a bioinformatics approach to systematically identify chimeric transcripts present in human mRNA and EST databases. Remarkably, in an oligodendroglioma EST library, they found an out-of-frame fusion involving CAMTA1 (no. 82 in their Supplementary Table 3), representing a chromosomal inversion whose likely functional consequence is to effectively inactivate one copy of CAMTA1 in that tumor.

We thank Dr. Nathan Ellis, Dr. Diane Tabarini, Miguel Ilzarbe, and Donna Wong of the Memorial Sloan-Kettering Cancer Center Core DNA Sequencing Facility for their help and Dr. Qiulu Pan for expert technical advice.