Identification by full-coverage array CGH of human DNA copy number increases relative to chimpanzee and gorilla

Abstract

Duplication of chromosomal segments and associated genes is thought to be a primary mechanism for generating evolutionary novelty. By comparative genome hybridization using a full-coverage (tiling) human BAC array with 79-kb resolution, we have identified 63 chromosomal segments, ranging in size from 0.65 to 1.3 Mb, that have inferred copy number increases in human relative to chimpanzee. These segments span 192 Ensembl genes, including 82 gene duplicates (41 reciprocal best BLAST matches). Synonymous and nonsynonymous substitution rates across these pairs provide evidence for general conservation of the amino acid sequence, consistent with the maintenance of function of both copies, and one case of putative positive selection for an uncharacterized gene. Surprisingly, the core histone genes H2A, H2B, H3, and H4 have been duplicated in the human lineage since our split with chimpanzee. The observation of increased copy number of a human cluster of core histone genes suggests that altered dosage, even of highly constrained genes, may be an important evolutionary mechanism.

Gene duplication has long been considered a primary mechanism of adaptive evolution (Ohno 1970). In theory, newly duplicated genes are redundant, and relaxed functional constraints allow acquisition of sequence changes in support of new functions and expression patterns. The importance of gene duplication in human evolution is supported by numerous studies that have documented DNA copy number differences between human and nonhuman primates. These studies have used diverse molecular approaches based on karyotyping (Yunis et al. 1980), physical clone maps (Fujiyama et al. 2002), partial coverage genomic arrays (Locke et al. 2003), cDNA arrays (Fortna et al. 2004), end-sequence profiling (Newman et al. 2005), draft genome sequence (Chimpanzee Sequencing and Analysis Consortium 2005), and high-quality sequence from chimp Chromosome 21 (orthologous to human Chromosome 22) (Watanabe et al. 2004). Careful alignment of accurately finished genome sequences from human and other primates promises to reveal DNA copy number differences at the highest possible resolution. Finished sequence for human is available, but the whole-genome shotgun (WGS) assembly for chimpanzee remains in draft form. The WGS approach is of proven value for rapidly and economically generating a full genome sequence, but it is clear that current approaches to assembling WGS data sets can underrepresent recently duplicated genome segments (She et al. 2004). Thus, to achieve a complete genome survey of DNA copy number in human versus chimpanzee, we have performed comparative genome hybridization (CGH) using the first full-coverage bacterial artificial chromosome (BAC) array of the human genome, which consists of 32,855 overlapping clones providing ∼79 kb average resolution. This is a direct approach and yields the greatest coverage at the highest level of resolution thus far achieved for comparison of these three species.

Results

Using full-coverage BAC array CGH, we executed a three-phased approach to identify segments of human genomic DNA that have likely been acquired since divergence from the common ancestor we share with chimpanzee. First, two samples of human genomic DNA (gDNA), one pooled from seven unrelated males and the other pooled from four unrelated females, were cohybridized to identify and exclude nodes on the array that gave aberrant ratios in a human-only comparison. Pooled DNAs were used in order to minimize the number of hybridization experiments, and to favor the detection of fixed rather than polymorphic copy number differences. Of the 31,842 mapped autosomal clones on the array, 212 showed aberrant ratios (>1.5 H-spread; see Methods) in the human-human comparison, and were excluded from further analysis. Next, we hybridized the human test DNA sample pooled from seven human males to a reference DNA sample comprised of DNA pooled from three unrelated male chimpanzees (Coriell Institute, Repository numbers NAO3448, NAO3450, NAO3452) (Fig. 1). These hybridizations were repeated under dye reversal, and a total of 1319 clones (855 increases, 464 decreases) were identified that consistently showed ratios that exceeded threshold in both dye orientations. As an added measure of stringency, we retained clones only if (1) they were confirmed by an equivalent copy number aberration in at least one additional overlapping clone, or (2) their location in the human reference genome sequence (NCBI_34) is supported by both their restriction digest pattern and BAC end sequence placement (Krzywinski et al. 2004). Under these criteria, a total of 585 clones gave elevated ratios in human relative to chimpanzee.

Figure 1.

Human chimpanzee DNA copy number ratio determined by full-coverage BAC array CGH.

We used gorilla as an outgroup to determine the most likely ancestral copy number state. By parsimony, human chromosomal segments showing an increased copy number ratio relative to both chimpanzee and gorilla most likely represent insertions specific to the human lineage. This is true regardless of whether the human genomic region containing the given segment is more similar to chimpanzee or gorilla. Note, however, that there are further caveats to the parsimony approach that must be considered. While the widely accepted species tree of hominoids places human and chimpanzee as a clade, with gorilla as an outgroup, there are regions of the genome that are incongruent with the species tree. For regions of the genome consistent with a human-gorilla clade, the assignment of a copy number increase to human is unaffected, that is, parsimony still favors a single event in the human branch over two independent events in the chimp and gorilla branches. However, a study by Chen and Li (2001) reported that 12 of 53 surveyed loci were more consistent with a chimp-gorilla clade than a chimp-human clade. For these regions only, parsimony is blind to whether there is a copy number increase in the human branch, versus a decrease in the branch leading from the human/chimp/gorilla node to the chimp/gorilla node. Thus, regarding the ancestral copy number state, our hybridization results will be ambiguous for ∼20% of the genome. In some regards, orangutan may be a more suitable outgroup, since the ratio of unresolved ancestral polymorphism to divergence is much lower because of the longer divergence time. However, a potential drawback in using orangutan as an outgroup in these experiments is that the arrays are spotted with human genomic clones, and hybridization becomes less reliable when more distant species are evaluated.

Thus, we proceeded with hybridization of the pooled human male test DNA sample to reference DNA from a single female gorilla (Coriell Institute, Repository number NGO5251). We decided to use human male test DNA rather than female test DNA for consistency with previous experiments. Because we had fewer chimpanzee and gorilla samples than human samples, there is some possibility that sites that are polymorphic in chimpanzee and gorilla have impacted our analysis. The fact that we restrict analysis to genomic segments where chimpanzee and gorilla copy number agree, relative to human, minimizes this impact. Of the 585 clones that had an elevated ratio in human relative to chimpanzee, 235 also gave elevated ratios relative to gorilla and therefore likely represent human-specific copy number increases. Presumably, the subset of clones that did not show elevated ratios relative to gorilla represent copy number decreases in chimpanzee relative to the ancestral state. Again, as an added measure of stringency, clones have been retained in the set of 235 only if they are confirmed by an equivalent copy number aberration in at least one additional overlapping clone, or their location in the human reference genome sequence (NCBI_34) is supported by both their restriction digest pattern and BAC-end sequence placement. These 231 clones collapse into 55 contiguous chromosomal segments (43 with multiple clones, and 12 singletons) with minimum, maximum, and average segment lengths of 65,252 bp, 1,133,633 bp, and 308,959 bp, respectively, and a cumulative genome footprint of 16,992,728 bp.

Separately, we evaluated ratios of clones located on the X and Y chromosomes. We identified a total of eight X chromosome (ChrX) and 28 Y chromosome (ChrY) clones that met the criteria of concordant dye-flip ratios and an equivalent copy number difference in at least one overlapping clone. Sex chromosome ratios from the female gorilla sample are not directly comparable to those from the male chimpanzee and human reference samples, thus for sex chromosome differences we are unable to infer human increase rather than chimp decrease. However, evaluation of duplicate segments within the human reference genome sequence (below) supports the notion that these are copy number increases in the human lineage. These eight ChrX clones and 28 ChrY clones collapse into two ChrX contigs and six ChrY contigs covering 415,787 bp and 1,190,263 bp, respectively, bringing the cumulative genomic footprint of all segments (autosomal plus sex chromosome) to 18,598,778 bp (Table 1). These segments are the basis of further analysis. While loss of genetic material on the human lineage is of considerable interest, here we consider only observed copy number increases. This is because copy number increases, as opposed to losses, can be readily validated through design of quantitative PCR experiments and through evaluation of signatures of duplication events in the reference human genome sequence as we describe below.

Table 1.

Position in the human genome (NCBI build 34) of 63 DNA segments with increased copy number relative to chimpanzee and gorilla, as determined by full-coverage BAC array CGH

Chromosome	Start	End	Size (kb)	Band
1	16,264,647	16,504,347	239,700	1p36.13
1	103,514,958	103,782,587	267,629	1p21.1
1	141,699,150	141,904,149	204,999	1q21.1
1	142,697,868	142,832,783	134,915	1q21.1
1	145,670,162	146,179,409	509,247	1q21.1–1q21.2
1	146,800,368	147,090,803	290,435	1q21.2
2	87,255,531	88,008,041	752,510	2p11.2
2	87,880,005	87,967,456	87,451	2p11.2
2	89,146,792	89,285,612	138,820	2p11.2
2	89,290,779	89,526,330	235,551	2p11.2
2	89,855,739	90,009,157	153,418	2p11.2
2	91,079,860	91,278,418	198,558	2p11.2
2	110,374,075	110,628,474	254,399	2q13
2	112,292,849	112,493,396	200,547	2q13
2	132,579,662	132,809,765	230,103	2q21.1–2q21.2
4	13,423	177,154	163,731	4p16.3
4	70,432,214	70,618,319	186,105	4q13.2
5	26,149,382	26,333,080	183,698	5p14.1
5	69,007,757	69,735,382	727,625	5q13.2
5	69,955,192	70,348,029	392,837	5q13.2
5	112,886,922	113,064,958	178,036	5q22.2
6	170,704,322	170,894,763	190,441	6q27
7	60,835,494	61,035,840	200,346	7q11.1
7	64,392,229	64,533,018	140,789	7q11.21
7	71,931,071	72,131,909	200,838	7q11.23
7	73,503,814	73,734,450	230,636	7q11.23
7	73,914,085	74,089,099	175,014	7q11.23
7	142,658,374	142,723,626	65,252	7q34
7	143,134,261	143,465,465	331,204	7q35
8	47,000,811	47,258,017	257,206	8q11.1
9	38,905,428	39,384,898	479,470	9p13.1
9	39,765,473	39,978,303	212,830	9p12
9	40,174,404	40,340,379	165,975	9p12
9	40,495,504	40,759,034	263,530	9p12
9	41,111,334	41,356,670	245,336	9p12
9	41,431,201	41,545,828	114,627	9p11.2
9	41,557,881	42,021,347	463,466	9p11.2
9	43,787,741	44,127,010	339,269	9p11.2
9	63,574,163	64,135,077	560,914	9q13
9	65,216,018	65,523,621	307,603	9q13
9	65,617,653	65,814,846	197,193	9q13–9q21.11
10	46,138,281	46,380,466	242,185	10q11.22
10	57,969,110	58,079,974	110,864	10q21.1
14	18,070,001	18,405,573	335,572	14q11.2
15	18,898,763	19,129,849	231,086	15q11.2
15	19,301,328	19,916,381	615,053	15q11.2
15	20,813,690	21,041,101	227,411	15q11.2
15	22,004,834	22,248,614	243,780	15q11.2
15	26,055,804	26,752,132	696,328	15q13.1
16	21,002,359	21,198,510	196,151	16p12.3
16	32,089,294	32,319,125	229,831	16p11.2
16	70,583,784	71,175,111	591,327	16q22.1–16q22.2
18	42,695,217	42,966,859	271,642	18q21.1
20	58,697,375	58,867,659	170,284	20q13.32–20q13.33
21	21,603,697	21,743,689	139,992	21q21.1
22	14,440,103	15,014,058	573,955	22q11.1
X	87,762,347	87,994,542	232,195	Xq21.31
X	88,109,770	88,293,362	183,592	Xq21.31
Y	8,909,830	9,017,798	107,968	Y
Y	9,598,790	9,717,440	118,650	Y
Y	19,570,010	19,871,468	301,458	Y
Y	21,103,620	21,264,863	161,243	Y
Y	25,095,852	25,388,816	292,964	Y
Y	27,429,589	27,637,569	207,980	Y

Since it is expected that genomic segments recently gained in the human lineage have originated through duplication of existing sequence, we evaluated the degree of overlap between segments identified by CGH and segments identified by in silico analysis (BLAST matches >1 kb long with >90% identity, as described in Krzywinski et al. 2004) of the reference human genome sequence (NCBI_34). Whereas coverage of the genome by in silico predicted segmental duplications is 5.2%, in silico predicted duplications covered 73.4% of the sequence represented in the 63 CGH-identified chromosomal segments, more than an order of magnitude enrichment. This is consistent with the notion that copy number gains as identified by CGH have arisen through segmental duplication. While the sequence identity between in silico defined duplicates can suggest an approximate duplication date, only phylogenetic analysis allows a time interval to be directly estimated for the duplication event. Of the total 8499 in silico defined duplicated segments, 176 (2.1%) intersect with the portion of the genome that appears from CGH to be have been gained specifically in the human lineage, suggesting that these duplications occurred in the past 4 to 6 million years (the divergence time of human and chimpanzee). It is important to note, however, that this is a minimal estimate for the proportion of recent duplicates, since segments smaller than the resolution of the BAC array (∼79 kb) will not have been detected by CGH.

A total of 192 non-pseudogene Ensembl genes were detected on the 63 duplicated segments. If these segments arose through segmental duplication, we would expect representation from paralogous genes within this set. The coding sequences of these genes were compared by reciprocal BLAST analysis (expect-value cutoff = 10^-10), which identified 41 strict paralogous gene pairs (82 genes total) (Table 2). For these genes, pairwise synonymous and nonsynonymous substitution rates were estimated for the aligned sequence using the codon substitution models of Yang and Nielsen (2000) as implemented in PAML (phylogenetic analysis by maximum likelihood) (Yang 1997). The mean d_N/d_S ratio of the gene set is 0.433, the median ratio is 0.388, and the highest observed ratio is 1.56. Likelihood ratio tests were applied to test the null hypothesis of equal rates of substitution (H₀: d_N = d_S) versus the alternative hypothesis (H_A: d_N ≠ d_S). For 15 of the 41 genes, the null hypothesis had a nominal P-value <5%, and in all but one of these significant cases, d_N/d_S < 1, implying that purifying selection had been removing deleterious amino acid variants, and that both copies of the gene must have retained function for at least a fraction of the time subsequent to the gene duplication. An extreme case of protein conservation since the time of gene duplication would be if d_N/d_S = 0 (i.e., no amino acid changes were tolerated), and in this case it might be appropriate to compare the null hypothesis to the alternative with d_N/d_S < 0.5. This null hypothesis arises because we allow for one homolog to be strictly conserved (d_N/d_S = 0) and the other to be strictly neutral (d_N/d_S = 1), thus a somewhat more liberal test of one of the pair showing excess divergence considers as a null hypothesis the average of d_N/d_S = 0.5 (Thornton and Long 2002). This test identified as significant three genes that had d_N/d_S > 0.5 (Table 2), but in no case have these genes been functionally characterized.

Table 2.

Strict paralogous gene pairs (reciprocal best BLAST matches) present on human DNA segments with increased copy number relative to chimpanzee and gorilla

Gene 1	Chr	Start	End	Description	Gene 2	Chr	Start	End	Description	dN	dS	dN/dS	P(1.0)	P(0.5)
ENSG00000186301	1	16,353,203	16,360,133	Similar to macrophage stimulating 1 (hepatocyte growth factor-like)	ENSG00000186715	1	16,462,486	16,468,660	No description	0.233	0.843	0.276	0.000	0.000
ENSG00000051415	1	103,423,056	103,655,958	α-Amylase, salivary precursor	ENSG00000174876	1	103,584,694	103,592,978	α-Amylase salivary precursor	0.015	0.061	0.245	0.000	0.044
ENSG00000132043	1	146,102,223	146,214,718	No description	ENSG00000168614	1	145,923,370	145,958,684	No description	0.268	0.260	1.032	0.694	0.000
ENSG00000182639	1	147,000,080	147,000,460	Histone H2B.J	ENSG00000184678	1	147,074,392	147,074,772	Histone H2B.Q	0.004	0.188	0.021	0.001	0.004
ENSG00000183598	1	147,001,408	147,001,818	H3 histone family 2; H3 histone family, member M	ENSG00000183702	1	147,040,763	147,042,418	H3 histone, family 2; H3 histone family, member M	0.000	0.418	0.001	0.000	0.000
ENSG00000183558	1	147,030,455	147,030,847	Histone H2A.O	ENSG00000183717	1	147,039,210	147,039,773	Histone H2A.O	0.000	0.001	0.387	0.758	0.811
ENSG00000182217	1	147,047,334	147,049,286	Histone H4	ENSG00000183941	1	147,020,824	147,022,776	Histone H4	0.001	0.000		0.797	0.743
ENSG00000169769	2	89,344,195	89,344,735	IG κ chain V-III region NG9 precursor	ENSG00000184817	2	87,540,174	87,540,735	No description	0.051	0.061	0.839	0.749	0.324
ENSG00000173756	2	89,423,308	89,423,634	IG κ chain V-II region GM607 precursor	ENSG00000182563	2	89,330,327	89,330,632	No description	0.092	0.237	0.388	0.014	0.500
ENSG00000131426	2	89,897,439	89,897,914	IG κ chain V-I region HK 101 precursor	ENSG00000163245	2	89,301,505	89,301,971	IG κ chain V-I region	0.027	0.029	0.903	0.888	0.403
ENSG00000172038	2	89,912,079	89,912,595	IG κ chain V-III region CLL precursor (rheumatoid factor)	ENSG00000186192	2	89,058,824	89,248,127	IG κ chain C region	0.045	0.098	0.459	0.116	0.858
ENSG00000163184	2	89,957,240	89,957,539	IG κ chain V-I region	ENSG00000186856	2	89,062,541	89,242,332	No description	0.000	0.021	0.001	0.040	0.105
ENSG00000169606	2	132,594,457	132,595,857	No description	ENSG00000172981	22	14,635,352	14,636,482	No description	0.031	0.165	0.191	0.000	0.001
ENSG00000154927	2	132,690,833	132,735,739	No description	ENSG00000186825	2	132,763,505	132,764,092	No description	0.000	0.006	0.001	0.118	0.210
ENSG00000177631	4	49,320	78,099	No description	ENSG00000182141	22	15,009,448	15,026,562	No description	0.327	0.582	0.562	0.000	0.003
ENSG00000184671	5	69,006,181	69,062,049	Baculoviral IAP repeat-containing protein 1 (neuronal apoptosis inhibitory protein)	ENSG00000185284	5	69,659,709	69,678,715	No description	0.036	0.113	0.318	0.000	0.051
ENSG00000182078	5	69,234,613	69,234,750	No description	ENSG00000186932	5	70,293,834	70,293,971	No description	0.000	0.001	0.001	0.766	0.874
ENSG00000183666	5	69,498,165	69,537,080	No description	ENSG00000183761	5	69,239,339	69,301,070	No description	0.000	0.001	0.000	0.682	0.832
ENSG00000180027	5	70,249,240	70,289,288	SMA3 protein	ENSG00000185759	5	69,997,497	70,038,832	SMA3 protein	0.000	0.001	0.410	0.730	0.813
ENSG00000174133	5	70,271,857	70,276,207	No description	ENSG00000186361	5	70,010,592	70,014,942	No description	0.039	0.026	1.514	0.001	0.000
ENSG00000170135	7	73,622,573	73,679,914	Transcription factor GTF2IRD2	ENSG00000174428	7	73,962,893	73,977,192	Transcription factor GTF2IRD2	0.052	0.099	0.524	0.002	0.820
ENSG00000123965	7	73,718,985	73,733,868	Postmeiotic segregation increased 2-like 5	ENSG00000135165	7	71,762,053	75,683,468	Zona pellucida sperm-binding protein 3 precursor	0.007	0.009	0.793	0.802	0.611
ENSG00000165178	7	73,984,535	73,999,836	No description	ENSG00000182487	7	72,046,778	72,062,076	No description	0.003	0.023	0.137	0.007	0.072
ENSG00000181881	7	143,339,071	143,443,973	No description	ENSG00000181926	7	143,383,314	143,389,573	No description	0.000	0.000	0.858	0.991	0.832
ENSG00000170341	7	143,406,977	143,407,891	Seven transmembrane helix receptor	ENSG00000170356	7	143,320,745	143,321,671	Seven transmembrane helix receptor	0.000	0.004	0.001	0.104	0.189
ENSG00000106714	9	39,063,817	39,278,466	Contactin associated protein-like 3 precursor (cell recognition molecule CASPR3)	ENSG00000154529	9	39,704,516	39,839,734	No description	0.009	0.020	0.422	0.229	0.809
ENSG00000147926	9	39,908,370	39,912,422	No description	ENSG00000182355	9	39,345,699	39,351,955	No description	0.009	0.006	1.560	0.295	0.005
ENSG00000179828	9	40,483,601	40,501,134	No description	ENSG00000182153	9	41,776,397	41,793,931	No description	0.000	0.003	0.001	0.154	0.261
ENSG00000170165	9	43,977,955	43,988,544	No description	ENSG00000176057	9	43,907,570	43,918,159	No description	0.000	0.003	0.001	0.147	0.251
ENSG00000170161	9	63,660,398	63,661,000	No description	ENSG00000186383	9	41,506,025	41,506,582	No description	0.015	0.055	0.269	0.091	0.420
ENSG00000176299	14	18,238,610	18,239,551	Seven transmembrane helix receptor	ENSG00000182974	15	19,866,208	19,867,149	Seven transmembrane helix receptor	0.029	0.028	1.034	0.942	0.097
ENSG00000176294	14	18,285,736	18,286,659	Seven transmembrane helix receptor	ENSG00000183706	15	19,880,105	19,881,055	Seven transmembrane helix receptor	0.077	0.191	0.402	0.000	0.356
ENSG00000175733	15	21,016,715	21,019,583	No description	ENSG00000184095	15	19,057,902	19,060,769	No description	0.083	0.091	0.913	0.756	0.025
ENSG00000128731	15	25,958,549	26,169,609	HECT domain and RLD 2	ENSG00000140181	15	20,829,990	20,883,368	No description	0.020	0.034	0.599	0.017	0.391
ENSG00000153684	15	26,233,151	26,236,568	No description	ENSG00000183629	15	26,495,524	26,498,941	No description	0.000	0.001	0.001	0.590	0.671
ENSG00000169861	16	32,106,704	32,113,996	No description	ENSG00000182414	16	32,009,310	32,099,878	IG heavy chain V-III region	0.305	0.533	0.572	0.036	0.611
ENSG00000183677	18	42,811,361	42,813,622	RNA polymerase II transcription factor SIII subunit A2 (elongin A2)	ENSG00000183791	18	42,806,560	42,808,200	RNA polymerase II transcription factor SIII subunit A3 (elongin A3)	0.070	0.085	0.825	0.381	0.018
ENSG00000130538	22	14,823,378	14,824,325	No description	ENSG00000186445	14	18,171,223	18,172,203	Seven transmembrane helix receptor	0.008	0.000		0.066	0.016
ENSG00000185912	Y	8,898,452	8,960,986	Testis-specific Y-encoded protein	ENSG00000187194	Y	9,000,204	9,001,613	No description	0.052	0.126	0.412	0.089	0.699
ENSG00000131007	Y	19,638,276	19,645,782	Transcript Y 9 protein	ENSG00000131009	Y	19,788,761	19,796,267	Transcript Y 9 protein	0.000	0.002	0.001	0.821	0.941
ENSG00000169953	Y	19,788,522	19,830,805	Heat shock transcription factor, Y-linked	ENSG00000172468	Y	19,603,741	19,646,021	Heat shock transcription factor, Y-linked	0.000	0.002	0.001	0.601	0.633

Gene identifiers, coordinates, and descriptions are from the Ensembl database (v. 27.35a) and are based on the NCBI_34 genome build. P-values for likelihood ratios are given for observed d_N/d_S compared to a null d_N/d_S of 1.0 or 0.5, respectively.

We selected two duplicated loci (AMY1A and CNTNAP3) for evaluation by an independent method (real-time quantitative PCR; Taqman). These loci were sequenced in our human, chimpanzee, and gorilla samples, and then primer/probe sets were designed to regions of sequence that were perfectly conserved between duplicates and among species. Of note, a third human amylase family member (AMY2B) was present on the duplicated segment that contained the two copies of AMY1A; thus the amylase primer probe sets were designed to a region of exact sequence identity among all three amylases. Results from these PCR assays (Fig. 2) verify increased copy number of these loci in human versus chimpanzee and gorilla.

Figure 2.

Relative copy number of the AMY1A and CNTNAP3 loci in three unrelated chimpanzees and a single gorilla compared to a single pooled sample of human gDNA.

Discussion

Using full-coverage BAC array CGH, we have identified 63 genomic segments with an increased hybridization ratio in human versus chimpanzee. Because these segments also show an increased hybridization ratio in human versus gorilla, the most parsimonious explanation is that these CGH-defined segments have been duplicated very recently in human evolutionary history, subsequent to our divergence from chimpanzee. This interpretation is supported by the high representation within these segments of in silico defined human segmental duplications, and the verification by real-time quantitative PCR of copy number differences at selected loci. However, the formal possibility remains that some subset of these CGH-defined segments has been independently lost in both chimpanzee and gorilla, rather than gained in humans. Owing to high sequence similarity among these three closely related primates and the substantial length of the CGH BAC probes (∼200 kb), it is exceedingly unlikely that sequence divergence is responsible for any observed differences. It must be considered that a portion of the genome does not represent the species tree but, rather, supports a chimp-gorilla clade over a chimp-human clade. For copy number differences in this portion of the genome, which remains to be accurately mapped, parsimony is not effective in assigning the ancestral copy number state. However, the fact that we have relied on gorilla as an outgroup should not have a significant impact on the results of the present study because we evoke parsimony only in the first data-filtering step of our analysis. Subsequent analysis is strictly focused on paralogous gene pairs within genomic segments with copy number alteration. Tandem duplication is a signature of DNA copy number increase, and provides a level of internal validation to our analysis. Furthermore, where we have done quantitative gene dosage analysis for further verification of human copy number gains (Fig. 2), the data have supported this interpretation.

Interestingly, we observe a substantial DNA copy number increase at chromosome 2q13 in human. This is the site of the telomeric fusion between chimpanzee chromosomes 12 and 13 in the human/chimp common ancestor that resulted in human chromosome 2 (Yunis et al. 1980). Furthermore, the expansions of human chromatin adjacent to the centromeres of chromosomes 1, 9, and 16 that were noted by Yunis et al. (1980) were also observed in the present study (Table 1). Fortna et al. (2004) previously used cDNA arrays for comparative analysis of hominoid genomes, and found a total of 25 relatively large DNA segments (11.1 Mb average length) that appeared to be duplicated in the human lineage. Of the 63 human copy number gains we measured in the present study, 30 segments have coordinates that agree with this previous study. The remaining 33 segments appear to be novel findings from our complete coverage array.

We analyzed the gene content of the 63 chromosomal segments with increased copy number in human. Genes within these regions were subjected to reciprocal BLAST analysis to find duplicated copies. Among the 41 high-confidence paralogous gene pairs we detected, the most highly represented gene family is immunoglobulin (IGK) genes, with five paralogous pairs. This is consistent with earlier whole-genome comparative analysis of, for example, fly and mosquito (Christophides et al. 2002; Hill et al. 2002; Zdobnov et al. 2002) or rat and mouse (Gibbs et al. 2004), where immune-related gene families have been found to be prone to expansion. The second most highly represented gene family is histones, with four paralogous pairs. In particular, the present study highlights human-specific duplications of members of the core histone minor cluster at chromosome 1q21 (Table 2). Histone octamers (comprised of two proteins each from core histone multimember families H2A, H2B, H3, and H4) form the core of the nucleosome, and are among the most evolutionarily conserved of all proteins, with substantial sequence conservation between humans and organisms as divergent as sea urchin (Grunstein et al. 1976). It is possible, given the relatively strict structural requirements of the nucleosome and exceptional sequence conservation across species, that variation in gene dosage becomes a viable alternative to sequence variation for adapting histone expression and function in accordance with selective pressures. While it is not clear what selective advantage might be conferred by many of the human duplicated genes highlighted in this study, of primary interest are genes with a potential role in nervous system development. An interesting candidate in this regard is ENSG00000106714 (Contactin-associated protein, CNTNAP3) and its uncharacterized paralog ENSG00000154529. CNTNAP3 is a member of the neurexin family of cell recognition molecules. Neurexins and their membrane-bound ligands (neuroligins) are thought to mediate interactions between neurons, including synapse formation. CNTNAP3, a member of the NCP subgroup of neurexins, is expressed throughout the human brain and is important in ion channel localization and neuron-glial interactions (Spiegel et al. 2002). Should further investigation of CNTNAP3 and its paralog verify non-pseudogene status, it will be important to evaluate what role these genes might have in synaptic function.

Under neutral evolution, coding mutations will be fixed at the same rate as silent mutations, giving a d_N/d_S ratio of 1. The median d_N/d_S ratio observed in our gene set was 0.388, which is consistent with net purifying selection acting on these recently duplicated genes. This observation is consistent with previous reports of reduced d_N/d_S ratios between paralogous genes in Drosophila (Thornton and Long 2002) and Arabidopsis (Zhang et al. 2002), and is indicative of continued function of both gene copies subsequent to the duplication event. While several genes in our set had d_N/d_S ratios >1, and a strict application of the PAML test of H₀: d_N/d_S = 1 versus H_A: d_N/d_S > 1 identified only one case of positive selection, a more liberal test that uses H₀: d_N/d_S = 1/2 (and an alternative hypothesis with d_N/d_S as a free parameter) identifies an additional two unannotated genes with weak support for significant positive selection. This latter test might, instead, be considered a test for constraint, however, as it assumes that the original functional gene copy tolerates zero nonsynonymous changes.

The d_N/d_S ratios reported here are average ratios for the aligned length of each protein pair. Identification and sequencing of the strict orthologs of these genes in chimpanzee and additional primates will allow evaluation of synonymous and nonsynonymous substitution rates in a site-specific and lineage-specific manner and will likely yield further insight into human adaptive evolution. Further exploration of genes and noncoding functional sequences within the boundaries of these variable segments will be helpful for elucidating the genetic basis of human-specific traits.

Methods

Comparative genomic hybridization

Hybridizations were done using the whole-genome SMRT array (Ishkanian et al. 2004), which consists of amplified MseI fragments from 32,433 tiled Human BACs (Krzywinski et al. 2004) spotted in triplicate on two aldehyde-coated slides, and gives an effective resolving power of 79 kb. Test and reference DNAs were digested with MseI, labeled by random priming with the fluorescent nucleotide analogs Cy5-dCTP and Cy3-dCTP, and purified using Sephadex G50. For each hybridization experiment, the test and reference DNAs were combined and denatured, and repetitive sequences were blocked by coincubation with denatured human Cot-1 DNA (Invitrogen). Repeat-blocked DNA was then hybridized at 45°C for 48 to 72 h. After hybridization, the slides were washed at 45°C for 15 min in 80% DIGEasy hybridization solution (Roche Scientific), 2× SSC (pH 7), followed by three washes of 5 min at room temperature with 0.1× SSC (pH 7), 0.1% SDS, four rinses of 30 sec each in 0.1× SSC (pH 7) at room temperature, a brief rinse with deionized, distilled water, and then dried. Imaging was done using a Packard Biosciences Scan Array Express instrument.

Microarray analysis

An open-source software package, called MIA, was developed and implemented to extract intensity ratios from the raw images. The analysis of CGH array images is divided into several broad steps: (1) addressing, which consists of finding the location of subarrays and individual tiles containing one and only one spot; (2) segmentation, which consists of identifying the pixels belonging to the spot within each tile; (3) extraction of spot and background intensities; and (4) normalization of data. These steps are not specific to CGH experiments, and the software can therefore be useful for any two-color microarray work such as the analysis of gene expression with spotted arrays. Similar to the approach of Yang et al. (2001), the addressing and segmentation steps are performed on a combined 8-bit image obtained by a square-root transformation, but the intensities are extracted from the original raw 16-bit images. Several mathematical techniques were applied in order to obtain a completely automated addressing procedure. The average spacing between spots is deduced by analyzing the Fourier transform of one-dimensional spectra projected in both the horizontal and vertical directions. The average spot size is obtained by granulometry, more precisely, by studying the effect of successive morphological openings with structural elements of increasing size (Soille 2003). Once the subarrays are located, individual tiles are first positioned on a regular two-dimensional square lattice, then an optional optimization can be performed for each tile. The Seeded Region Growing algorithm (Adams and Bischof 1994) was used in the segmentation process of each tile. A square seed (by default of size two by two pixels) associated with the spot is positioned at the center of intensity of the tile. The pixels on the edge of the tile serve as background seeds unless their intensity falls within the top 10% of that tile, and the pixels with intensities above that threshold form the seeds for artifacts. Therefore, following the application of the Seeded Region Growing algorithm, each pixel within the tile is assigned as being part of the spot, part of the background, or part of an image artifact. There are no geometrical constraints applied to the shape of the spot except that it is obviously constrained to the limits of the tile as determined at the addressing stage. Spot intensities are extracted from the raw 16-bit images. The spot pixels as determined in the segmentation phase are averaged to extract the spot intensity. Similar to the work of Yang et al. (2002), the background intensity is obtained by probing a morphological opening obtained with a large structural element (default of 2.5 times the average spot spacing). Intensity ratios were normalized with the help of the robust LOESS regression on the so-called M-A plot, where M = log₂(I₁/I₂) and A = log₂[(I₁ × I₂)^1/2], with I₁ and I₂ being the background subtracted intensities of the spot in the two images.

We selected individual thresholds for each array using the same type of calculations used to compute box-and-whisker plots (Tukey 1977). In a box-and-whisker plot, the difference between the upper (H2) and lower (H1) hinges is called the H-spread. The definition of hinge is similar to that of quartile and therefore H-spread is similar to the interquartile range. The upper (lower) whisker is located 1.5 × the H-spread above (below) the upper (lower) hinge, and values outside the whiskers are considered extreme. For each individual array, the thresholds for DNA copy number aberrations were set at the whisker levels. A MySQL relational database called CGHdb was built and populated with all ratio and ancillary data. Results from a human female (test) versus human male (reference) hybridization allowed estimation of true-positive, false-negative, and false-positive rates based on sex chromosome copy number. A total of 1134 of the 1430 X-chromosome clones had increased copy number, and 157 of the 196 Y chromosome clones had decreased copy number in the female test DNA sample. These values give a true-positive rate of 79.2% for increases and 80.1% for decreases, with corresponding false-negative rates of 20.8% for increases and 19.9% for decreases. Of the 30,216 autosomal clones, 665 (2.2%) showed increased copy number, and 592 (2.0%) showed decreased copy number. It is not possible to determine if these autosomal copy number differences reflect true positives or false positives.

Quantitative PCR

Primer/probe sets for two separate test genes (AMY1A and CNTNAP3) were designed using the repeat masked reference human genome sequence (NCBI_34; April 2003 release; http://genome.ucsc.edu/) (Table 1). Primers and probes were selected in regions of exact match between the test gene and its top paralog, as defined below (Identification of Paralogs). The SNAP25 gene, previously established to be single copy in the human genome (Bailey et al. 2002), was selected as a reference gene. The target sequences for test and reference primers and probes were verified in our human, chimp, and gorilla samples by direct sequencing. To determine relative copy number, 10 ng of genomic DNA was assayed in triplicate in 20-μL reactions containing 1× final concentration TaqMan Universal Master Mix (ABI part number 4304437), 200 nM each primer and probe, and 10 ng of template DNA. Each experiment was performed using a 384-well optical PCR plate and the ABI 7900HT machine with default cycling conditions. Copy number of the test locus in chimp or gorilla versus human was defined as 2^-ΔΔC_T, where ΔC_T is the difference in threshold cycles for the test and reference loci.

Identification of paralogs

We sought to identify paralogous genes within the 55 recently duplicated human genomic regions. cDNA sequences for 206 Ensembl genes from these regions were collected with the Ensembl API v.27_35a. Using BioPerl v1.4 (Stajich et al. 2002), all sequences were transformed into peptides. Each gene's best partner within this set was determined by using the BLAST algorithm (Altschul et al. 1997) (E-value 10^-10, ungapped alignments) against the peptide library. Once a gene's best partner was identified, the BLAST algorithm was rerun with the best hit as the query sequence. If, in the second iteration, the best BLAST hit was the original query sequence, then this gene pair was deemed paralogous. High-scoring segment pairs (HSPs) from the aligned regions were extracted in-frame in nucleotide form for subsequent analysis of substitution rates.